Geoffy — Full text corpus (llms-full.txt) Site: https://geoffy.ai Generated: 2026-09-16 Sections: 12 marketing pages, 25 blog posts, 8 whitepapers, 39 documentation pages, 3 sector benchmark pages For agent guidance, see https://geoffy.ai/agents.md For the discovery file, see https://geoffy.ai/llms.txt ================================================================================ # Protocols make you findable. Geoffy makes you chosen. URL: https://geoffy.ai Last updated: 2026-07-21 ================================================================================ ## Protocols make you findable. Geoffy makes you chosen. When shoppers ask ChatGPT, Perplexity, Claude or Gemini what to buy, the assistant names two or three products — and ignores the rest. Geoffy is the system that gets your products named, and keeps them there. Live today on Shopify and WordPress / WooCommerce. The problem ## AI assistants are already recommending products. Are yours in the answer? Product discovery is moving from search results into AI conversations. The assistant doesn't return ten blue links — it gives one answer, names a couple of brands, and the buying decision starts there. The brand that gets named wins the consideration. The brand that doesn't never finds out it lost. Traditional SEO measures Google rankings. It doesn't tell you whether an AI names you, why it chose someone else, or what to do about it. That's a different discipline — and it needs different infrastructure. One system, three capabilities ## Optimise your store. Monitor your visibility. Influence the sources AI trusts. - Optimise (available now) - Your product pages, engineered so AI engines can read and cite them accurately. One widget on each product page — structured answers across every depth of buyer intent — with a matching structured-data layer underneath and drift control keeping the two in step. Every claim it publishes is verified against your own product data first — engines cite facts, not adjectives. No cloaking. No hidden text. Nothing a human shopper doesn't also see. - Monitor (early access) - Geoffy asks the AI engines the real questions your customers ask, on a schedule, and shows you who gets named, in what order, which sources the answer came from — and whether each gap is fixable on your own pages or has to be earned off-site. - Influence (early access) - For the gaps you can't fix on your own domain, Geoffy prepares the material to earn citations on the sources AI actually trusts — retailer listing briefs, outreach packs, authority briefs. Grounded in the exact sources the AI cited. You place it; we never fake anything. ## Monitor finds the gap. Optimise or Influence fixes it. Monitor shows the movement. Monitor finds the gap, routes it to Optimise or Influence, the fix earns the citation, and Monitor shows the movement. Competitors sell one piece of this. Geoffy closes the loop. ## Monitor and Influence are in early access. We're rolling them out with a small group of customers so the measurement earns trust before it scales. Optimise is generally available today. How this is different ## SEO platforms ship recommendations. Geoffy ships the fix. Visibility trackers tell you where you show up and stop. SEO suites produce audit lists that end with “send to developer”. Agencies do the work — at £3,000–6,000 a month, without an instrument underneath. Geoffy is productised GEO: the system does the engineering, on your own domain, and shows its working. - Engineers your pages - Geoffy yes | monitoring tools no | SEO suites no | agency retainer yes - Measures AI answers - Geoffy yes | monitoring tools yes | SEO suites partial | agency retainer partial - Routes gaps to fixes - Geoffy yes | monitoring tools no | SEO suites no | agency retainer yes - Prices like software - Geoffy yes | monitoring tools yes | SEO suites yes | agency retainer no - First-party on your domain - Geoffy yes | monitoring tools no | SEO suites no | agency retainer yes What you actually get ## What you actually get - The Geoffy widget on every product page - Answers structured across Broad, Mid and Ultra buyer intent, including who each product is for — and who it isn't. That honesty is what makes engines trust the rest. - Verified claims only - Every published statement checked against your product data; hype adjectives banned; your distinctive claims (vegan, UK-made, sugar-free) engineered into every layer engines read. - A structured-data mirror - Says exactly what the page says. Never more. - Entity clarity - Engines know precisely which brand and which product you are — consistent naming and authoritative identity links, so your citations never leak to a similarly named rival. - Drift control - When your catalogue changes, Geoffy keeps page, widget and data in step. - The full machine-readable layer - Llms.txt, agents.md, buying guides on your domain, and a GEO Score per product so you can watch coverage improve. Free audit ## ChatGPT is recommending products in your category right now. See if it's recommending yours. Drop your domain. Get your GEO Score and the exact questions where you're invisible in AI search. Platforms ## Works with your commerce platform. Live today on Shopify and WordPress / WooCommerce. Magento is on the roadmap; BigCommerce and Adobe Commerce are planned. - Shopify - available now - WordPress - available now - Magento - planned - BigCommerce - planned - Adobe Commerce - planned ## Built for modern commerce teams. - Shopify stores - Get your products named in the answers your customers already ask for. - Agencies - Run productised GEO across a client portfolio, with an instrument underneath. - WordPress and WooCommerce stores - First-party AI discovery on your own domain. ## Questions teams ask before they start - Do I need to replatform to use Geoffy? :: No. Geoffy works on your existing catalogue and your existing product pages. - Can I start with Shopify or WordPress today? :: Yes. Both are live today. - Does Geoffy publish hidden AI-only content? :: No. Nothing is published that a human shopper doesn't also see. - What happens as products change? :: Drift control keeps the page, the widget and the structured data in step. - Which AI assistants does this work with? :: The engines your customers already use, including ChatGPT, Perplexity, Claude and Gemini. - Is there an agency programme? :: Yes — a multi-client platform with wholesale pricing. ## Being findable is table stakes. Being chosen is the game. Run your free GEO Score, or talk to us about the full loop across Optimise, Monitor and Influence. ================================================================================ # The engineering behind being chosen | Geoffy URL: https://geoffy.ai/product Last updated: 2026-07-21 ================================================================================ ## The engineering behind being chosen. Geoffy is coherence infrastructure for AI commerce: it engineers your product pages so AI engines cite them accurately, measures what the engines actually say, and prepares the off-site work you can't do from your own domain. One product. Three capabilities. One loop. Optimise (available now) ## One widget on the canonical product page. Not a parallel site of “AI pages”. Early GEO tooling — ours included, in its first version — created separate AI-facing pages at scale. The industry learnt the same lesson search learnt twenty years ago: doorway patterns don't hold. Geoffy's architecture puts everything on the page that already earns your trust signals: the canonical product page. - The widget - A structured block rendered on each product page, holding answer-ready content across three depths of buyer intent — Broad (“I need a coat”), Mid (“a waterproof, warm coat”), Ultra (brand, spec, use case, thresholds). Visible in the DOM, because ChatGPT, Claude and Perplexity read the raw HTML at fetch time and don't execute your JavaScript. - The structured-data mirror - JSON-LD that matches the visible page exactly, including variant handling done properly (ProductGroup wrapping for variable products). Schema feeds the engines' upstream understanding; the visible layer is what gets quoted. You need both, and they must agree. - Claim verification - Every claim, verified. Geoffy doesn't write marketing copy; it verifies claims. Every statement published about a product is checked against your own product data before it ships. Vague sells (“premium”, “durable”, implied benefits) are banned outright. What survives is the set of distinctive, factual, comparable claims — vegan, sugar-free, UK-made, 20g protein — engineered into the visible page, the structured data and the machine-readable files. Those are the claims AI engines select on, and unverifiable ones are exactly what they've learnt to distrust. - Entity disambiguation - Consistent brand and product identity everywhere engines look, plus authoritative identity links (Wikipedia/Wikidata, company registries). AI engines confuse similar brands constantly; Geoffy makes you unambiguous. - The machine-readable layer - Llms.txt, llms-full.txt and agents.md, served correctly from your own domain and kept in step with the catalogue automatically. - Agent-ready pages - Semantic structure and a clean accessibility tree, so shopping agents parsing your pages get the same clear answer a human would. - Drift control - Catalogues change daily. Geoffy's parity checks catch the moment page, widget and structured data disagree, and repair it. - Coherence audit - Before anything publishes, Geoffy checks the new content against what's already on the page. Conflicts get flagged, not shipped. - Fast publication - On publish, caches are invalidated first, then search infrastructure is pinged, so engines see the new version in hours, not weeks. - Buying guides - Category-level answer pages served on your own domain, covering the comparison questions assistants answer daily. Monitor (early access) ## Measurement you can act on — and trust. Monitor asks the engines the real, high-intent questions your customers ask (“best sustainable leather boots under £150?”) on a schedule, and extracts the answer's anatomy: who's named and in what order, which sources were cited, and how the answer was produced. That last part is the diagnostic that matters — the pathway split. If the engine retrieved your page and still didn't name you, that's fixable in Optimise. If it answered from training memory, the fix is off-site authority — that's Influence. If we can't tell, we say “unknown” rather than guess. Monitor is built as an honest instrument: results are labelled estimated or live-verified, a measured zero is never confused with “couldn't measure”, and movement is reported as correlation, not causation. Measurement you'd bet a budget on has to earn it. - Grounded in your real distribution - Monitor knows which retailers and marketplaces actually stock you, confirmed by you rather than guessed from a directory, so it never routes work toward shelves you're not on — and can tell you when a stockist you are on is winning citations you're not getting. - An evidence log behind every number - Each result keeps the raw engine response it came from. When Monitor says a competitor was cited, you can read the actual answer. Measurement you can audit, not a score you have to take on faith. Influence (early access) ## The gaps you can't fix on your own domain. Some answers are decided off your site — on the retailer listings, review sites, videos and reference pages the engines cite. Influence generates the ready-to-place material to earn those citations: retailer listing briefs, outreach and pitch packs, video metadata packs, authority briefs — each grounded in the exact sources the AI cited for that question. It ends at the draft. You, or your agency, do the placing. Owned and earned only — no fake reviews, no sockpuppets, no bought placement. That line is both an ethical position and a competitive one. The loop ## Measure → diagnose → fix → prove. Monitor finds a gap and routes it — on-page gaps to Optimise, off-page gaps to Influence — with the context attached: the question asked, who got cited, which source won. When the fix lands and the citation follows, Monitor shows the movement. That closed circuit is the product. Everything else in the category sells one arc of it. Transparency ## Nothing hidden. That's the point. No cloaking. No hidden prompts. No bot-only content. No invented attributes — if we don't know a spec, the field stays empty. Structured data always matches the visible page. AI engines are getting better at detecting manipulation every month; the durable strategy is being the brand whose data deserves the citation. This isn't a compliance posture bolted on afterwards — verification is the mechanism. Engines cite what they can trust; we make you the easiest brand in the category to trust. - No cloaking - No hidden prompts - No invented attributes - Verified claims only Platforms ## Works with your commerce platform. Live today on Shopify and WordPress / WooCommerce. Magento is on the roadmap; BigCommerce and Adobe Commerce are planned. ## Get your products engineered for the answer. Install on your platform, or start with the free GEO Score to see where you stand today. ================================================================================ # Agency-grade GEO. Software prices. | Geoffy pricing URL: https://geoffy.ai/pricing Last updated: 2026-07-21 ================================================================================ ## Agency-grade GEO. Software prices. A GEO agency retainer runs £3,000–6,000 a month. Geoffy productises that work — the on-page engineering, the measurement, the off-site briefs — as one system with three capabilities. Pick a bundle, or build your own. All plans are billed monthly. No long contracts. Bundles ## Three closed-loop bundles. Bundles include priority access to Monitor and Influence as they roll out. Each bundle is a saving on the three capabilities bought separately. Monitor allowances are high-intent buyer questions (“best sustainable leather boots under £150?”), not tracked keywords. A hundred strategic questions beat a thousand junk ones. - Foundational Loop :: £199/mo :: For brands just discovering they're invisible in AI answers. Optimise: up to 250 products engineered | Monitor (early access): 30 high-intent questions, monthly, ChatGPT | Influence (early access): 3 citation briefs + 1 outreach pack per month - Grow Loop :: £499/mo :: For scaling D2C brands defending a category position. Optimise: up to 1,000 products engineered | Monitor (early access): 100 questions, weekly, ChatGPT + Perplexity | Influence (early access): 15 briefs + 5 packs per month - Dominate Loop :: £1,499/mo :: For category leaders and large catalogues. Optimise: up to 5,000 products engineered | Monitor (early access): 300 questions, weekly + on-demand, all engines | Influence (early access): unlimited briefs and packs Build your own ## Start with one capability. Add the rest when you're ready. Every capability is a valid entry point and upgrades independently. Your price is simply the sum of what you pick. Example: on Grow Loop with a 2,000-product catalogue? Upgrade just Optimise to Advanced and pay the difference. - Optimise - Products engineered :: Basic £99/mo | Standard £249/mo | Advanced £749/mo - Monitor - Questions & cadence (early access) :: Basic £29/mo | Standard £99/mo | Advanced £299/mo - Influence - Briefs & packs (early access) :: Basic £79/mo | Standard £199/mo | Advanced £499/mo ## Included in everything First-party publishing on your own domain · verified claims only — checked against your product data before publish · visible-content architecture with a matching structured-data mirror · entity disambiguation · llms.txt + agents.md served and maintained · drift control and coherence checks · GEO Score for every product · no cloaking, ever. ## Not sure where you stand? Start with the free audit. Your GEO Score across the major engines, per product, free. It's the same measurement the paid system uses — just a snapshot instead of a schedule. ## Platform support - Shopify - available now - WordPress - available now - Magento - planned - BigCommerce - planned - Adobe Commerce - planned ## Running GEO for clients? There's a partner programme with a multi-client platform and wholesale pricing. ## Enterprise, or beyond 5,000 SKUs Multi-brand, marketplaces, custom SLAs, API access — by conversation. ## Pricing questions - Why “talk to us” instead of a checkout? Monitor and Influence are in early access; we onboard the loop with you rather than dropping you in a dashboard. Self-serve is coming. - Is there a trial? The free GEO Score is the try-before-you-buy: real measurement, your real products. - What counts as a product? An active SKU we engineer and keep in step. Variants handled properly, not double-counted. - Can I change bundles? Monthly billing; move up, down or modular whenever. - Do you gate which AI engines? Monitor cadence and engine coverage scale with level. Optimise output is readable by every engine at every level — it's on your own pages. ## Pick a bundle, or start with the free score. ================================================================================ # What is GEO? Generative Engine Optimisation explained | Geoffy URL: https://geoffy.ai/what-is-geo Last updated: 2026-07-21 ================================================================================ ## Generative Engine Optimisation (GEO) The discipline of structuring product information so AI assistants can interpret it, trust it, and recommend it. If SEO was about ranking in a list of links, GEO is about being the answer. The shift ## The naming moment is the new front door of ecommerce. Search engines rank pages; the shopper clicks and decides. AI assistants generate answers; the shortlist is decided before any click happens. When someone asks “what's the best protein powder for marathon training?”, the assistant names two or three products. That naming moment is the new front door of ecommerce — and most product catalogues were never structured for it. SEO vs AEO vs GEO ## Three disciplines, three targets. The overlap is real: much of GEO is built on disciplines SEO people already know — structured data, crawlability, content quality. The difference is what you're optimising for: a ranked list versus a synthesised answer. - SEO - Gets your pages ranked in search results. Still matters; different game. - AEO - Gets facts extracted into answer boxes. - GEO - Gets your products included and cited in AI-generated answers and recommendations: the consideration set plus the citation. It's the union of being in the shortlist and being the referenced source. Reference Engineering ## The practice, not a tactic. Reference Engineering is Geoffy's practice within GEO: engineering the canonical product page so an AI engine can read it, trust it, and cite it. Concretely, that means one structured widget on the product page carrying answer-ready content across the full range of buyer intent, a structured-data layer that mirrors the visible page exactly, and controls that keep the two coherent as the catalogue changes. Underneath it all, one discipline: only verified, distinctive, comparable claims — because those are the statements AI engines can check, trust and repeat. What it explicitly is not: separate “AI pages”, hidden content for bots, or attribute-stuffing. The industry tried the parallel-page pattern; it doesn't hold, for the same reason doorway pages didn't hold in search. The two layers ## AI engines meet your products twice. Upstream, structured data feeds the knowledge systems engines build their world model from — slow-moving, compounding over weeks and months. At retrieval, when a user asks a question and the engine fetches your page in real time, it reads the visible content. ChatGPT, Claude and Perplexity don't execute your JavaScript at that moment, so anything that only appears after client-side rendering isn't there as far as they're concerned. And when Ahrefs tracked 1,885 pages adding JSON-LD in May 2026, AI citations moved by a statistically insignificant margin — schema alone is not the unlock at the moment of answer. Do only schema and you're invisible at the moment of answer. Do only visible content and you never enter the world model properly. GEO is doing both — and keeping them in agreement. Intent Tiers ## Broad, Mid and Ultra — blocks inside one widget. Customers ask at different depths: Broad (“I'm looking for a coat”), Mid (“a waterproof, warm coat”), Ultra (“Gore-Tex, fur-lined, for the Himalayas, under £300”). Coverage across all three — as blocks inside one widget on the product page — is how a product stays in the answer whichever way the question is phrased. ## GEO questions - Is GEO replacing SEO? :: No. Different surfaces, different mechanics. Your SEO keeps working; GEO covers the surface SEO can't see. - Do GEO pages replace my product pages? :: No — that's the point. The widget lives on your product page. No parallel site. - Does GEO involve hidden content? :: Not here. Everything Geoffy ships is visible on the page and mirrored in the structured data. Hidden bot-content is both against platform policies and, more practically, ignored at retrieval. - Which platforms? :: Live today on Shopify and WordPress / WooCommerce. Magento is on the roadmap; BigCommerce and Adobe Commerce are planned. - How do I know if I have a GEO problem? :: Ask an AI assistant to recommend products in your category. If you're not in the answer, you have a GEO problem. Or run the free GEO Score and see it measured. ## See the mechanism, or measure yourself against it. ================================================================================ # Your Shopify catalogue, recommended by AI | Geoffy for Shopify URL: https://geoffy.ai/solutions/shopify Last updated: 2026-07-21 ================================================================================ ## Your Shopify catalogue, recommended by AI Geoffy installs like any Shopify app, renders through the official theme app extension system, and engineers every product page for AI discovery — without touching your theme files. Built as a Shopify citizen ## Built as a Shopify citizen. - Two-click theme integration - Add the Geoffy block in the theme editor via Shopify's app-block mechanism. No Liquid edits, no code pasted into your theme, removable instantly. - Metafields, not hacks - Generated content lives in metafields; the widget renders from them. Your theme stays yours; theme updates don't break anything. - Verified claims, engineered in - Every statement Geoffy publishes about a product is checked against your own catalogue data first, with vague sells banned, so what reaches the page is the set of distinctive, factual claims engines actually select on. - Variants done properly - Variable products get correct group-level structured data, so engines understand the product family rather than seeing twelve confusing near-duplicates. - Root discovery files - The machine-readable discovery files AI crawlers look for, served correctly from your domain, with content quality that goes beyond the platform defaults. - Buying guides on your domain - Category-level answer pages served through the app on your own storefront, covering the “best X for Y” questions assistants answer all day. - Fast propagation - Publishes invalidate caches first, then ping index infrastructure, so changes reach the engines in hours. ## Why Shopify stores specifically Assistants increasingly answer shopping questions with specific product recommendations, and well-structured Shopify stores are heavily represented in those citations. The mechanics are on your side: clean catalogue data, fast pages, a proper app architecture. What's missing on most stores is the answer-ready layer — which is exactly the piece Geoffy adds. ## What it costs Bundles from £199/mo, modular options from £29/mo. ## Install once. Answer-ready from there. ================================================================================ # AI discovery infrastructure for agencies | Geoffy URL: https://geoffy.ai/solutions/agencies Last updated: 2026-07-21 ================================================================================ ## AI discovery infrastructure for agencies Add productised GEO to your service line. Geoffy does the engineering and the measurement; you own the strategy, the reporting and the client. The margin arithmetic ## Senior problem becomes junior problem. A £3–4k/month GEO retainer normally burns senior strategist hours — the person who has to work out entity data, structured-data doctrine and citation diagnostics. On Geoffy, the same deliverables are generated by the system and reviewed by a junior. Senior problem becomes junior problem; the retainer's margin transforms. You pay a platform fee and wholesale module pricing; you bill the client for strategy and execution. ## The programme - Platform fee + wholesale - Agency platform at £299/mo (multi-client dashboard, per-client switching, multi-user access) plus module pricing at wholesale rates you mark up or rebill. Exact terms by conversation. - The prospecting hook - Run a visibility scan on a prospective client and walk in with a branded “here's how invisible you are in AI answers” report. It wins pitches. - Do-it-with-us, not white-label-everything - Influence generates the briefs and pitches; your team places them and bills for it. White-label reporting available where it makes sense. - Eligibility - Agencies managing 3+ ecommerce clients. ## Agency features - Multi-client workspace - Per-client analytics - Discovery reporting - Optional white-label reporting - Multi-user access ## Early access Monitor and Influence are in early access; agency partners are first in the queue — the prospecting scan is an early-access benefit worth having. ## Build a recurring, margin-positive GEO service line. ================================================================================ # AI product discovery for WordPress and WooCommerce | Geoffy URL: https://geoffy.ai/solutions/wordpress Last updated: 2026-07-21 ================================================================================ ## AI product discovery for WordPress and WooCommerce A proper WordPress plugin — not a script tag — that engineers your WooCommerce catalogue for AI recommendation, and plays correctly with your caching layer and your SEO stack. Built as a WordPress citizen ## Built as a WordPress citizen. - Standard plugin install - Install, connect WooCommerce, done. No theme edits, no replatforming. - WooCommerce variants handled properly - Variable products get group-level structured data with correct variant relationships, the thing most schema plugins get wrong, and the thing that makes engines misread product families. - Verified claims, engineered in - Every statement Geoffy publishes about a product is checked against your own catalogue data first, with vague sells banned, so what reaches the page is the set of distinctive, factual claims engines actually select on. - Caching-aware publishing - Geoffy hooks the major caching layers (WP Rocket, LiteSpeed, W3 Total Cache, Cloudflare, host-level caches) so a publish invalidates the cache before index infrastructure is pinged. Order matters: ping first and the engines re-read your stale page. - Respects your editorial content - Product structured data goes on products; your posts and pages keep their native markup. No accidental “everything is a Product” pollution. - Buying guides on your domain - Category answer pages served through the plugin on your own site. - Coexists with your SEO plugin - Geoffy adds the AI-answer layer; Yoast/RankMath and friends keep doing their job. ## Why WooCommerce stores specifically WooCommerce stores have a structural advantage most owners don't use: full control of the page. No platform rendering restrictions, no locked templates. That control is exactly what the visible-content layer of GEO needs — Geoffy just does the engineering for you. ## What it costs Bundles from £199/mo, modular options from £29/mo. ## Install the plugin. Keep your stack. ================================================================================ # About Geoffy — Coherence Infrastructure for AI Commerce URL: https://geoffy.ai/about Last updated: 2026-07-21 ================================================================================ ## About Geoffy. Geoffy is a focused team building coherence infrastructure for AI commerce. We help ecommerce brands become the reference answer in AI shopping conversations. ## Reference Engineering. Reference Engineering is the practice we apply. It structures product information so AI systems understand it, trust it, and recommend it. Distinct from SEO. SEO ranks pages; Reference Engineering structures references so AI assistants treat the brand as the reference answer. ## How customers actually ask People ask AI assistants at three depths. Broad: “I'm looking for a coat.” Mid: “a waterproof, warm coat.” Ultra: “Gore-Tex, fur-lined, for the Himalayas, under £300.” A product needs to be legible at every depth to stay in the answer — which is why Geoffy structures every product page's content across all three, in one place, on the page that already carries your trust signals. No parallel pages, no hidden layers: one coherent product page a machine can quote confidently. ## What makes Geoffy different - Built for AI discovery, not retrofitted SEO - The unit of work is the answer, not the ranking. - Engineering plus evidence - We don't just structure your pages; we measure what the engines actually say and route every gap to a fix (measurement in early access). - Transparent by design - Visible content, matching structured data, no invented attributes, no cloaking. If we don't know a spec, the field stays empty. - First-party, always - Everything publishes on your domain, under your control. Your discovery infrastructure shouldn't belong to a marketplace. ## Operator. - Operated by Geoffy Limited (UK Companies House #17295484). - Based in the United Kingdom. - Contact: hello@geoffy.ai ================================================================================ # Connect your store. Geoffy does the engineering. | How it works URL: https://geoffy.ai/how-it-works Last updated: 2026-07-21 ================================================================================ ## Connect your store. Geoffy does the engineering. No replatforming, no theme surgery, no six-week implementation project. ## The install, step by step - Connect — Shopify: install the app from the App Store and add the Geoffy block in your theme editor, Shopify's official block mechanism, a couple of clicks. WordPress: install the plugin, connect WooCommerce. - Geoffy reads your catalogue — products, variants, attributes, existing content. This is the raw material; nothing is invented on top of it. Unknown spec? The field stays empty. - Review the generated content — for each product: the widget content across Broad, Mid and Ultra intent, plus the structured-data mirror. Every claim has been verified against your catalogue data before you see it — anything unverifiable was never written. You review, you approve, you stay the editor. - Publish — the widget renders on your product pages, on your domain. Caches are invalidated first, then the index infrastructure is pinged, so engines pick up the new version in hours, not weeks. - It stays correct on its own — price change, spec update, product retired: drift checks catch the difference and keep page, widget and structured data in step. Coherence checks run before every republish. This is the part that would otherwise be a standing agency task. Honest mechanics ## “No theme file edits” — here's what actually happens. On Shopify, Geoffy adds a theme app extension block via Shopify's supported injection mechanism — nothing pasted into your theme files, removable in one click. On WordPress, it's a standard plugin that plays properly with your caching layer and SEO plugins. We'd rather explain the mechanism than hand-wave it. Then: find out if it's working ## Optimise is the work. Monitor is the proof. Once your pages are engineered, the question is whether the engines start choosing you. Monitor — in early access — asks the engines your customers' real questions on a schedule and shows who's named, which sources won, and whether each remaining gap is an on-page fix or an off-site one. ## Platforms Live today on Shopify and WordPress / WooCommerce. Magento is on the roadmap; BigCommerce and Adobe Commerce are planned. ## Connect your store and let the engineering run. ================================================================================ # How AI assistants actually choose products | Geoffy URL: https://geoffy.ai/ai-product-discovery Last updated: 2026-07-21 ================================================================================ ## How AI assistants actually choose products Not magic, not a black box — a retrieval mechanism with observable rules. Understand the mechanism and the optimisation strategy writes itself. Synthesis, not retrieval ## Answers are synthesised, not retrieved. A search engine returns pages and lets you judge. An AI assistant fetches from multiple sources, weighs them by trust and consistency, and generates one answer with a handful of products named. - The shortlist happens before the click - If you're not in the synthesis, you're not in consideration. - Sources compete inside a single answer - Your product page competes with retailer listings, review sites and videos to be the cited source about your own product. - Consistency is weighting - When sources disagree about your product, the engine trusts all of them less, including you. The retrieval-time read ## At retrieval, engines read what a human would see. When an engine fetches a page in real time to answer a question, it reads the visible HTML. ChatGPT, Claude and Perplexity don't execute JavaScript at that moment — searchVIU's crawler analysis classifies GPTBot, ClaudeBot and PerplexityBot as unable to render it — so content that only exists after client-side rendering may as well not be there. This explains a lot of failed AEO tooling. When Ahrefs tracked 1,885 pages that added JSON-LD, against 4,000 matched controls, AI citations moved +2.4% on Google AI Mode and +2.2% on ChatGPT — both statistically indistinguishable from zero. Schema on its own didn't move the answer. Hidden AI-only content is structurally useless for the same reason: the engine isn't reading the hidden layer at answer time. It's why the Geoffy widget renders in the DOM, visibly, on the canonical page. The upstream layer ## Schema for registration; visible HTML for extraction. Retrieval is half the story. Structured data feeds the knowledge systems engines consult to build their candidate pool — which brands and products even get considered for retrieval. That layer moves slowly and compounds, and it's precisely the use Ahrefs' own authors preserve for schema: knowledge graphs, entity recognition, rich results. Schema for registration; visible HTML for extraction; both mandatory; both in agreement. That's the two-layer model in one sentence. Where citations come from ## The citation substrate isn't all your website. The sources AI engines cite for ecommerce answers are measurable, and they're not all your website: video platforms, retailer listings, marketplaces and community discussion sit alongside brand-owned pages. - Brand-owned pages win a large share of consideration-stage citations - BrightEdge's May 2026 analysis found brand-owned commercial pages take the largest share of consideration-stage citations in every one of the eight industries it studied, ranging from 42% to 79%. That's the slot the widget serves directly. - The rest is off-site - Which is why measurement without off-site remediation is a dead end, and why Influence exists. The coherence penalty ## Contradictions are expensive. Old prices in cached fragments, spec conflicts between description and data sheet, stale variant info — each one erodes machine trust in everything else the page says. When sources disagree about your product, including your own page disagreeing with itself, the engine has no way to tell which version is true, and the safe move is to cite someone else. Coherence isn't housekeeping; it's a ranking factor in the answer economy. This is why drift control and pre-publish coherence checks are core Geoffy machinery, not features. The same logic extends to claims. An engine can corroborate “vegan” or “20g protein per serving” across sources; it can do nothing with “premium quality”. Verifiable, distinctive claims are citable; adjectives are noise — and pages built on noise lose the citation to pages built on facts. Intent coverage ## Intent coverage is the real game. Customers phrase the same need a thousand ways at three depths — Broad, Mid, Ultra. Engines answer at whichever depth they're asked. A product visible at every depth, from one coherent page, stays in the answer across the whole question space. That's what the widget's three intent blocks are for. Where Geoffy fits ## The infrastructure between your catalogue and the answer engines. The widget as the on-page extraction surface, verified-claims engineering so what it publishes can be corroborated, the structured-data mirror as upstream registration, entity disambiguation so engines know exactly which brand you are, and drift and coherence control as the trust guard — plus, in early access, Monitor to measure what the engines actually say and Influence to earn the citations that live off your domain. ## See the product, or measure yourself first. ================================================================================ # Security & transparency - Geoffy URL: https://geoffy.ai/security Last updated: 2026-03-05 ================================================================================ ## Security and transparency. Geoffy is parity-first. What users see is what bots see. No cloaking. No hidden content. No sketchy tracking. ## Parity-first by default. - Visible content matches structured data. - No content shown only to crawlers. - No hidden prompt blocks. ## First-party publishing. GEO pages are published on your own domain. That keeps indexing, trust and ownership clean. ## Transparent outputs. - JSON-LD - /page.json parity endpoints - Sitemaps - llms.txt ## Want the details? See our legal and policy pages for full terms and processing commitments. ================================================================================ # Legal - Geoffy URL: https://geoffy.ai/legal Last updated: 2026-03-05 ================================================================================ ## Legal. Our policies and terms for using Geoffy. - Terms of Service - Privacy Policy - Cookie Policy - Data Processing Addendum - Acceptable Use Policy - Billing & Refunds - Security & Compliance ================================================================================ # AI visibility for footwear and trainer brands (Geoffy Sector Benchmark) URL: https://geoffy.ai/sectors/footwear Last updated: 2026-09-14 ================================================================================ Get AI to understand the facts that make your footwear worth recommending. 90 answers · 10 buying questions · 3 AI engines · 3 runs per question · UK · 14 September 2026 Footwear and trainers: 20 brands. Named = mentioned anywhere in the answer. Recommended = put forward as an option. Own site cited = at least one citation in the answer pointed to the brand's own domain. Brand | Named | Recommended | Own site cited - Loake: named 29, recommended 19, own site cited 21 - Crockett & Jones: named 22, recommended 20, own site cited 16 - Cheaney: named 21, recommended 17, own site cited 7 - Clarks: named 15, recommended 14, own site cited 9 - size?: named 15, recommended 11, own site cited 9 - Grenson: named 15, recommended 8, own site cited 4 - Tricker's: named 14, recommended 13, own site cited 3 - New Balance: named 14, recommended 8, own site cited 8 - Offspring: named 13, recommended 12, own site cited 3 - Vivobarefoot: named 13, recommended 10, own site cited 5 - Solovair: named 12, recommended 11, own site cited 5 - END.: named 12, recommended 8, own site cited 0 - Nike: named 12, recommended 2, own site cited 9 - Footpatrol: named 10, recommended 10, own site cited 7 - Dr. Martens: named 10, recommended 3, own site cited 1 - Schuh: named 10, recommended 2, own site cited 2 - Hotter: named 7, recommended 7, own site cited 1 - Kick Game: named 7, recommended 6, own site cited 1 - Footasylum: named 1, recommended 1, own site cited 0 - Oliver Sweeney: named 0, recommended 0, own site cited 0 ================================================================================ # AI visibility for garden and outdoor furniture brands (Geoffy Sector Benchmark) URL: https://geoffy.ai/sectors/garden-furniture Last updated: 2026-09-14 ================================================================================ Give AI the material, dimensions, guarantees and product facts it needs to recommend your furniture. 90 answers · 10 buying questions · 3 AI engines · 3 runs per question · UK · 14 September 2026 Garden and outdoor furniture: 20 brands. Named = mentioned anywhere in the answer. Recommended = put forward as an option. Own site cited = at least one citation in the answer pointed to the brand's own domain. Brand | Named | Recommended | Own site cited - Bramblecrest: named 32, recommended 30, own site cited 22 - Kettler: named 30, recommended 25, own site cited 10 - Hartman: named 26, recommended 22, own site cited 9 - John Lewis: named 23, recommended 12, own site cited 13 - Oxley's: named 16, recommended 16, own site cited 25 - Alexander Rose: named 15, recommended 11, own site cited 2 - Lazy Susan: named 14, recommended 14, own site cited 29 - Bridgman: named 12, recommended 11, own site cited 8 - 4 Seasons Outdoor: named 11, recommended 11, own site cited 0 - Barlow Tyrie: named 10, recommended 9, own site cited 2 - Harbour Lifestyle: named 6, recommended 5, own site cited 17 - Maze: named 5, recommended 4, own site cited 5 - Cox & Cox: named 4, recommended 3, own site cited 3 - Dunelm: named 3, recommended 3, own site cited 3 - Garden Trading: named 2, recommended 2, own site cited 1 - Dobbies: named 1, recommended 1, own site cited 1 - VonHaus: named 1, recommended 1, own site cited 0 - Moda Furnishings: named 0, recommended 0, own site cited 0 - Rowlinson: named 0, recommended 0, own site cited 0 - Danetti: named 0, recommended 0, own site cited 0 ================================================================================ # AI visibility for multi-brand menswear and streetwear retailers (Geoffy Sector Benchmark) URL: https://geoffy.ai/sectors/menswear-retail Last updated: 2026-09-14 ================================================================================ Give AI a reason to recommend your shop, not just the brands you stock. 90 answers · 10 buying questions · 3 AI engines · 3 runs per question · UK · 14 September 2026 UK multi-brand menswear and streetwear retailers: 16 retailers. Named = mentioned anywhere in the answer. Recommended = put forward as an option. Own site cited = at least one citation in the answer pointed to the retailer's own domain. Retailer | Named | Recommended | Own site cited - END.: named 58, recommended 50, own site cited 34 - Flannels: named 29, recommended 18, own site cited 12 - Sevenstore: named 23, recommended 20, own site cited 19 - Peggs & Son: named 18, recommended 17, own site cited 9 - Stuarts London: named 15, recommended 12, own site cited 12 - Mr Porter: named 15, recommended 12, own site cited 6 - Mainline Menswear: named 15, recommended 8, own site cited 12 - Hip: named 11, recommended 8, own site cited 0 - Footasylum: named 3, recommended 3, own site cited 1 - Philip Browne: named 1, recommended 1, own site cited 0 - Cho: named 1, recommended 1, own site cited 0 - Coggles: named 1, recommended 0, own site cited 1 - Woodhouse: named 0, recommended 0, own site cited 0 - Terraces: named 0, recommended 0, own site cited 0 - 80s Casual Classics: named 0, recommended 0, own site cited 0 - John Anthony: named 0, recommended 0, own site cited 0 ================================================================================ # Your Returns Data Is a List of the Facts Your Product Pages Left Out URL: https://geoffy.ai/blog/returns-data-is-a-content-brief Last updated: 2026-08-19 ================================================================================ Returns get treated as an operations problem. Reduce the rate, speed up the refund, work out which suppliers cause the most trouble. All sensible, and all of it looks backwards at a transaction that already went wrong. There is a second reading, and almost nobody uses it. A return is a record of a buyer who made a decision with incomplete information. They read the page, they formed an expectation, the product did not match it. Something they needed to know was not on the page, or was there and did not register. Which makes your returns log a list of the facts your product pages left out. Written by buyers. Product by product. In their words. ## What the return reason actually tells you Take the most common codes on a fashion or homeware store: size too large, size too small, not as described, wrong item, defective, colour, style, unwanted. An ops team reads "size too small" as a sizing problem and sends it to merchandising. Fair enough. But read the same code as a content signal and it says something more useful: this page did not carry the measurement, the fit note or the comparison a buyer needed to get the size right first time. The information exists somewhere in your business. It did not make it onto the page. "Not as described" is blunter still. That is a buyer telling you the page and the product disagreed. It is the single most direct statement of a content failure available anywhere in ecommerce, and it arrives already attached to a specific SKU. Now ask what an AI assistant does with that same page. An assistant answering "will this fit a standard 60cm cabinet" or "is this warm enough for winter walking" has to find the fact in the page it retrieved. If the fact is missing, the assistant does one of two things. It stays quiet about your product and names one that does carry the detail. Or it infers, and describes your product in terms you never wrote. The gap that produced the return and the gap that keeps you out of the answer are the same gap. One costs you a refund and a restock. The other costs you a mention you never knew was available. ## Be honest about how thin the native data is This is where most articles would tell you your returns data is an untapped goldmine. It is not, and overselling it would ruin the argument. On Shopify, the returns reason is a short closed list. The ReturnReason enum offers colour, defective, not as described, size too large, size too small, style, unwanted, wrong item, other and unknown. That is it. Free text is limited too: returnReasonNote caps at 255 characters and the customer's own note at 300. A closed list of ten codes and a couple of short text fields is not a rich corpus. Say so plainly, because a brand that opens its returns export expecting essays will close it again in five minutes. Some returns platforms are more generous. Loop Returns exposes a free-text return_reason through its warehouse reporting endpoint, so brands running Loop have real sentences rather than codes. Worth knowing which side of that line you are on before you plan any work. ## Thin still beats absent Here is why the thin signal is worth reading anyway. It names the product. Not a category, not a segment. This SKU, this failure, this buyer. Almost nothing else in your content workflow operates at that resolution. It names the failure mode. "Size too small" on one product and "not as described" on another are different content jobs, and the code tells you which one you have before you open the page. And it is a record no competitor holds. That is the part that matters for AI discovery. Your competitor can read your product page, your category, your reviews and your competitors' pages. They cannot read your returns log. A page written from it carries something theirs cannot. Ten codes at SKU resolution, aggregated across a season, will tell you more about which pages are failing than any keyword tool, because a keyword tool describes a market and a returns log describes your customers meeting your product. ## Turning it into pages The workflow is unglamorous. Export returns for the last two quarters with the reason code and the SKU. Group by product, then by reason. Ignore anything with one or two returns; you want products where the same reason repeats, because repetition is the signal that the page is at fault rather than the individual buyer. For each repeat, write down the fact that would have prevented it. A returned garment marked "size too small" three times needs the actual measurements and a comparison to a familiar reference, not a size chart link. A product returned as "not as described" needs the specific attribute the description over-promised, corrected. Then put that fact on the page as visible text. Not in a size chart popup, not in a downloadable spec sheet, not in a tab that loads on click. Assistants read what is served in the page. A fact that appears only after a JavaScript interaction is a fact they will not see. If the same missing fact turns up across a whole range, it belongs at collection level as well as on each product. ## What this does not prove Two limits worth stating. Nobody has shown that adding these facts makes an assistant recommend you. There is no study on real product data and commercial assistants, and anyone claiming a percentage lift from it is inventing one. The argument here is narrower: a page that answers a question a real buyer got wrong is more useful to a machine than one that does not, and the fact came from a record only you hold. And returns data has a survivorship problem. It only tells you about buyers who bought. The shopper who read the page, could not find the measurement and left is invisible in it. Your on-site search log is closer to that person, which is why Search Console cannot see your AI discovery problem argues for reading zero-result queries alongside this. ## The test to apply first Before you write any of these pages, run one check on the brief. Could a competitor with access only to public sources have written this page? For a page built from your returns log, the answer is no, and that is the whole point. For most of the content in most ecommerce calendars, the answer is yes. A page that fails that test may still be worth publishing. It will not be worth citing. For why understandable beats impressive when a machine is choosing, see AI recommends what it can understand. For the difference between depth and coherence, see SEO rewards depth, AI rewards coherence. ================================================================================ # Search Console Cannot See Your AI Discovery Problem URL: https://geoffy.ai/blog/search-console-cannot-see-ai-discovery Last updated: 2026-08-19 ================================================================================ Open almost any guide to getting your products mentioned by AI assistants and you will find the same first step: export your Search Console queries and build from there. It is a reasonable instinct. Search Console is free, already connected, and for twenty years the closest thing ecommerce had to a list of what buyers want. The habit is well earned. It is also, for this particular job, the wrong tool. Not because the data is bad, but because of what the data is. ## Search Console records clicks, not questions Search Console reports on queries that produced an impression or a click in Google Search. That is its scope, and it does the job honestly. An AI assistant does not work that way. When a shopper types "which of these will actually survive a dishwasher" into ChatGPT, that string is submitted to OpenAI. It is not a Google query. It produces no impression in your property, no click, no row in your export. Whatever the assistant then says about your product, whether it names you, ignores you or describes you inaccurately, happens entirely outside the reporting surface you are looking at. So the gap is not one of coverage or sampling. Search Console is not under-reporting AI demand. It has no mechanism to report it at all. That leaves a specific failure mode. If your content brief is built from a Search Console export, you are writing pages against the phrasing of people who used Google, in the past, and who behaved the way Google search behaviour rewards: short, keyword-shaped, stripped of context. The people asking assistants write in full sentences with conditions attached. Different input, different demand, different page. ## The overlap used to be big enough to ignore this For a while you could argue the two demand curves were close enough that a Google-shaped brief covered both. That argument is getting weaker, and there is now a number on it. Ahrefs studied 863,000 keyword SERPs and four million AI Overview URLs, published 2 March 2026. Only 38% of AI Overview citations came from pages ranking in the top ten. A year earlier the figure was 76%. Read that carefully, because it is easy to over-claim. It covers Google's own AI Overviews, not ChatGPT or Perplexity, and it describes where citations come from rather than what gets bought. But the direction is hard to argue with. The set of pages an AI answer draws on has come apart from the set of pages that rank. Optimising for the second is no longer a reliable way to land in the first. Which means the tool that tells you how you rank is losing its claim to be the tool that tells you what to write. ## The substitute is already on your own site Here is the part that tends to be missed. You do not need access to OpenAI's logs to see how your buyers phrase things. You have a search box. On-site search queries are prompts. They are typed by real buyers, in their own words, about your actual catalogue, on a property you own. Nobody has to grant you access and no vendor sits in the middle. The valuable subset is narrower still: the searches that returned nothing. A zero-result search is a buyer telling you, in their own language, about a thing they expected you to have or to explain, and could not find. Sometimes that means a genuine gap in the range. Very often it means the product is right there and the page does not use the words the buyer used — no mention of the material, the compatibility, or the threshold they cared about. That is not a merchandising problem. It is a page that fails to answer a question a real person asked, which is the same failure that keeps you out of an AI answer. On Shopify you can get at this two ways. The Search & Discovery app has a report called "Searches with no results", which covers the last thirty days. For a raw ongoing stream, the Web Pixels API emits a search_submitted event carrying event.data.searchResult.query alongside the variants that came back, so an empty productVariants array is a zero-result query, captured as it happens. Neither needs a new vendor. Both are live today. ## What this is not Three honest limits, because a brief built on a false premise is no better than one built on the wrong export. Your search box only hears from people who already reached your site. It tells you nothing about the shopper who asked an assistant, got three brands, and never arrived. On-site search corrects the phrasing problem. It does not solve attribution, and nobody should sell it as if it does. Volume is thin. A store doing modest traffic might see a few dozen zero-result queries a month, and the Search & Discovery report only looks back thirty days. This is qualitative material. Treat one query as a signal worth reading rather than a datapoint worth counting. And there is no study showing that fixing your zero-result queries makes an assistant recommend you. The argument here is upstream of that: a page written from a record only you hold cannot be produced by a competitor working from public sources. Whether that page then earns a citation depends on a lot of things this post does not claim to settle. ## What to do this week Pull your zero-result searches for the last thirty days. Read them rather than counting them. Sort them into two piles. One is genuine range gaps, which is a buying conversation. The other, usually the larger one, is products you already sell, described in words your buyers do not use. Every item in that second pile is a page brief, already written for you, in the buyer's phrasing, sourced from a record no competitor can see. Then ask the question that matters before you write anything: could a competitor with access only to public sources have produced this page? If the answer is yes, the page may still be worth publishing. It will not be worth citing. For what is and is not measurable once those pages are live, see measuring AI discovery. For why this channel stays quiet when you are losing, see the discovery channel that never tells you. ================================================================================ # Could a Competitor With Only Public Data Have Written This Page? URL: https://geoffy.ai/blog/the-counterfactual-competitor-test Last updated: 2026-08-19 ================================================================================ Every quality gate in content marketing runs at the end. You write the page, then something scores it: a readability tool, a similarity check, an AI-detection pass, an editor. All of those score the text. None score where the text came from. So a page can be well written, original by every similarity measure, technically clean, and still be a page four competitors will publish this quarter, because all five of you worked from the same sources. There is a cheaper gate, and it runs before anything is written. > Could a competitor with access only to public sources have produced this page? That is the whole test. One question, applied to the brief. ## Why the source matters more than the text Brand-owned pages are the majority of what AI assistants cite. Profound analysed 11.84 billion citations across eight models between April and July 2026, covering 3.02 million domains, and found roughly 57% of citations go to company-operated web properties. Social and UGC platforms sit far lower, and vary a lot by engine: Google AI Overviews cites social at 15.3%, while Microsoft Copilot uses a social source about once in every 29 citations. The intuitive version of this story, that AI prefers Reddit and reviews to brand sites, is not what the largest dataset says. That is good news and a problem at once. Good, because your own pages are the surface that gets cited. A problem, because so is everyone else's. Four inputs sit behind most ecommerce content briefs: the existing product page, Search Console, competitor pages, and general category knowledge. Every one is available to every competitor, and to every AI writing tool any of them points at their catalogue. That is not a quality problem. You can write beautifully from public inputs. It is a structural one. The same source material produces convergent pages, and a convergent page gives an assistant no reason to pick you over four others saying the same thing. ## Running the test Take the brief before it becomes a page and answer four questions. Source. Name the record this page draws on. If the honest answer is "the category" or "what competitors cover", the source is public. Counterfactual. Could a competitor working only from public sources produce this page? If yes, it will not differentiate you. Specificity. Does the page carry a fact, a figure or a sentence that exists only in your records? Destination. Is this at the right level, product, collection or site-wide? A page pitched at the wrong level fails even with a good source behind it. Question two is the one that does the work. The others explain the answer. ## Four worked examples A collection page for "waterproof walking boots". Source: category knowledge and a look at what three competitors did. Counterfactual: any of them could write it, and two already have. It fails. Re-brief it against the questions your own customers ask about waterproofing, drawn from your search log, support queue and returns reasons, and it becomes a page only you could write. A product FAQ block generated from the specification sheet. Source: your own product data, which sounds private and mostly is not. Your spec sheet is on your public page and probably your distributors' too. It fails, but only just. What rescues it is the questions: if the FAQ answers what buyers actually asked rather than what the spec implies, the source changes and the page passes. A buying guide comparing your range. Source: your catalogue, all public. It fails. It passes once it carries the thing no competitor holds: which products get returned against each other and why, or which comparison your support team makes every week. A page built from your zero-result search queries. Source: buyers typing into your search box and finding nothing. Not public, not purchasable, not inferable. It passes at the first question. The pattern is consistent. Public inputs produce pages that are fine and interchangeable. First-party records produce pages that are hard to reproduce. ## What to do with a page that fails Do not delete it. A page that fails the counterfactual is usually still worth having. Category pages need to exist and buying guides get used. The test does not sort content into keep and bin. It sorts content into cite-worthy and merely present. A page that fails is re-briefed against a first-party source, not thrown away. Often that means keeping the structure and replacing the evidence: same buying guide, comparison points now drawn from your returns log rather than a competitor's table. If a page genuinely has no first-party source available, publish it anyway and hold your expectations at the right level. It will do a job. It will not be the page that gets you named. ## The evidence on generic content, stated honestly There is a version of this argument that overreaches, and it is worth marking the line. Ahrefs studied around 331,000 pages across 100,000 SERPs in July 2026 and found pages with high AI-generated content signals received two to three times fewer impressions and about nine percentage points lower indexation. A real finding, pointing the right way. It is also correlational, and it relies on a proprietary detector. It does not show that generic content performs worse than publishing nothing. No study shows that, in AI retrieval or in search. And it is not evidence that Google penalises AI-generated content. The helpful content system was folded into core ranking in March 2024, and Google's spam policy targets intent to manipulate rankings rather than the use of AI. The honest claim is narrower and still strong enough: generic content is out-competed and under-indexed. It does not get you punished. It gets you passed over. ## Where to use it The test runs in about a minute per brief. Put it where a page gets commissioned, not where it gets reviewed, because by review time you have paid for the writing. If you run content through an agency it also gives you a clean instruction that does not require reviewing drafts line by line: name the first-party record behind each brief. Briefs that cannot name one get re-scoped before anyone writes. For how to choose between the tools that claim to measure any of this, see how to choose an AI visibility tool. For where the first-party records actually live, see Search Console cannot see your AI discovery problem and your returns data is a content brief. ================================================================================ # Measuring AI discovery: what to track, and what to refuse to claim URL: https://geoffy.ai/blog/measuring-ai-discovery-what-to-track Last updated: 2026-07-22 ================================================================================ Every ecommerce team asking about GEO eventually asks the same question: how do I know if it's working? It's the right question, and most of the industry answers it badly. So let's do this properly: what you can measure, what you can't, and how to tell an honest measurement system from a confident-sounding one. ## The problem is actually quite simple When a shopper asks an AI assistant for a recommendation, the assistant names a few products and moves on. No impression data reaches you. No search console. No rank tracker. The most important new surface in product discovery ships with no analytics. So measurement has to be reconstructed from the outside: ask the engines the questions your customers ask, record the answers, and extract the structure from them. That's straightforward to do badly and hard to do well — and the difference is worth understanding before you trust any number. ## What's actually worth tracking 1. Presence and rank. For a defined set of high-intent buyer questions — "best sustainable leather boots under £150?", not vague keywords — are you named? In what position? Naming order matters: assistants front-load their confidence. 2. Share of voice. Across your question set, how often are you named versus the competitors who keep appearing? The competitor list the engines produce is itself intelligence — it rarely matches the competitor list in your head. 3. Citations — who the engine trusted. When an answer cites sources, which won? Your product page, a retailer listing, a review site, a video? This tells you where the answer is actually decided, and it's frequently not where you're spending effort. 4. The pathway — the diagnostic almost everyone skips. There are two very different ways an engine produces an answer. It can retrieve — fetch pages live and synthesise from them. Or it can answer parametrically — from what it absorbed in training. The distinction decides your entire remediation strategy. Retrieved-but-not-chosen: your page was read and lost — fix the page. Parametric: the engine's memory of your category doesn't include you — no page edit fixes that; you need authority on the sources engines learn from. Same gap, opposite fixes. Measurement that doesn't tell you the pathway tells you that you lost, never why. 5. Movement against a control. AI answers drift on their own — models update, competitors act. If you change nothing and visibility moves anyway, that's the noise floor. Movement only means something measured against a baseline of untouched questions and benchmark brands. ## What an honest instrument refuses to do The uncomfortable part: this is probabilistic measurement of non-deterministic systems. The same question can produce different answers an hour apart. Any system claiming certainty here is overclaiming — and you're going to spend budget on these numbers, so overclaiming isn't a cosmetic sin. Four behaviours to demand from anything you use — including ours: - Zero vs unknown. "We measured, you weren't named" and "we couldn't measure this" are different facts. A system that renders both as zero is fabricating data. - Estimated vs verified, labelled. Some results come from APIs (fast, broad, approximate), some from checking the live product surface (slower, truer). You should always know which you're looking at. - Correlation, not causation. "You changed X and visibility rose" is a correlation claim. Genuine causal proof needs holdouts and time. A system that says "we improved your visibility by 40%" without a control is marketing, not measurement. - Abstention. Sometimes the honest answer is "we don't know yet". A system that never says this is guessing somewhere. We built Geoffy Monitor around exactly these rules — permanent holdout, benchmark drift basket, zero-vs-unknown discipline, labelled fidelity. Not because it makes the numbers more impressive; because it makes them safe to act on. Monitor is in early access with a small group of customers now. Request early access → ## Start smaller than you think You don't need a thousand tracked prompts. You need thirty good questions — the ones with a buyer behind them — measured consistently, with the pathway split, against a baseline. That beats a wall of vanity dashboards every time. And if you just want to know where you stand today: the free GEO Score runs this measurement once, on your real products. It's the same instrument, as a snapshot. Get your free GEO Score → ================================================================================ # The Anatomy of a GEO-Optimised Store URL: https://geoffy.ai/blog/anatomy-of-a-geo-optimised-store Last updated: 2026-07-16 ================================================================================ Most merchant websites are built for humans, and to some extent for Google. Pretty product pages, plenty of marketing content. An AI crawler doesn't care about any of that. To a machine deciding whether it can stake a recommendation on your catalogue, most of it is noise. So what does the destination look like — a store that AI engines can read, trust and cite? Here's the anatomy, piece by piece. Use it as a checklist. ## Discovery files at the root — mostly already there This part of the anatomy has changed, and it changed in the merchant's favour. Shopify now serves llms.txt, llms-full.txt and agents.md at the root of production stores by default, with agents.md declaring itself the canonical agent-facing file and the other two mirroring it. The platform settled the protocol argument; you are unlikely to be creating these files. Which leaves two jobs that are actually yours. - robots.txt — advertise the sitemap so any crawler hitting the root knows where to start, and allow the citation bots you actually want reading you. Blocking the wrong agent here quietly removes you from the answer. - agents.md — fill it. The default template is byte-identical across stores, with only the shop name and domain substituted in: no product authority, no brand voice, no curated answers. A clothing brand, a supplements brand and a homewares brand all ship the same file on day one. Whatever your platform serves, these files must agree with each other and with the visible site. A discovery layer that contradicts itself is worse than none. And note what they are: signposts to your real pages. They are not a place to publish a second version of your catalogue. ## A catalogue feed on your own domain Mirroring the XML sitemap: stable, predictable URLs listing every product with last-updated timestamps. Machines reward predictability; a feed that lives on your own domain keeps the authority yours. ## Products that answer at every depth of intent Shoppers ask at three depths — broad ("I need a new pair of slip-ons"), mid ("waterproof leather slip-ons for travel") and ultra-specific ("premium slip-ons, vegan-leather upper, arch support"). A GEO-optimised product page carries content for each depth: an overview paragraph, structured attributes, honest pricing and availability, usage guidance, FAQs, breadcrumbs, and the full JSON-LD an engine needs to ground a citation. Those depths are blocks on the product page, not destinations of their own. One page, answering the question however precisely it's asked. ## Not a parallel set of pages It is worth saying plainly what is not in the anatomy, because an earlier version of this post recommended it. A separate machine-readable surface — a /llm/{product}.md mirror of every product, or a set of AI-facing intent pages sitting alongside the real ones — looks efficient and is not. It splits your catalogue into two versions that immediately begin to disagree, it moves content away from the URL that carries your links, reviews and trust signals, and it asks a model to cite a page no human ever sees. That is the doorway pattern, and it does not hold. Everything else in this anatomy lands on the canonical product page. One page, engineered so a machine can read it — not a second site built for machines. ## Enriched schema on the human-facing page itself Most platforms emit some structured data by default. The GEO-optimised version extends it — correct offers data, review and rating markup where it exists, FAQ blocks, discovery attributes — and, critically, keeps it identical to what the visible page says. A crawler arriving directly at the product URL must get the same picture as one that came in via llms.txt. Schema that drifts from the page is a trust-destroyer. ## Collection pages that mirror it all at category level Where products share intent, the collection page you already publish carries the same kind of structured data and grounding content at group level — so an engine asked "what's the best vegan ketchup in the UK?" can land on a coherent category surface, not just one product. ## A monitoring layer over the top None of this is set-and-forget. Catalogues drift: prices change, products retire, a rewrite quietly contradicts a spec sheet. The final piece is high-frequency checking — findings with severity, tied to the specific page or element that triggered them, rolled into a coherence score with history so you can watch it move. Plus external tracking: querying the major AI engines on a schedule against the prompts that matter for your brand, recording whether you're cited, whether competitors are cited instead, and where the content gaps sit. ## The point That's the destination. What your platform lets you ship determines the path — Shopify and WooCommerce each have their own routes — but it doesn't change what the picture looks like. Every piece serves the same goal: when an AI reads your store, it finds one consistent, structured, checkable story about what you sell and who it's for. Building and maintaining all of this by hand is possible. Doing it across a full catalogue, and keeping it in sync as the catalogue changes, is what Geoffy is for. Optimise — the engineering layer — is generally available today on Shopify and WordPress / WooCommerce; Monitor and Influence are in early access. See which pieces your store already has — and which are missing: get your free GEO Score. ================================================================================ # The Discovery Channel That Never Tells You When You're Losing URL: https://geoffy.ai/blog/the-discovery-channel-that-never-tells-you-youre-losing Last updated: 2026-07-08 ================================================================================ Here's an experiment worth ten minutes of any founder's week. Take a genuine hero SKU — the one that pays the wages — and ask three AI assistants the question a real buyer would type. Nothing clever. For a supplements brand: "what's a good magnesium supplement for sleep and recovery?" Put it to ChatGPT, Perplexity and Gemini, one after the other. When we ran exactly this for one of our customers' bestsellers, the result was: three assistants, three different answers, and the product showing up in exactly one of them. That alone is worth sitting with. If this were Google, a result appearing on one engine and missing on two others would be a five-alarm fire — someone would be in a meeting about it by lunch. Because it happened inside a chat window, nobody at the brand knew it had happened at all. ## The follow-up is where it gets worse Then do the thing most people skip: ask the follow-up a picky buyer actually asks. What dosage? What form — citrate, glycinate, oxide? Is it third-party tested? The answers degrade. Not dramatically wrong, which would at least be easy to catch. Just vaguer — and in places subtly off: a dosage lifted from an old blog post, a "tested" claim stitched together from a marketplace listing that no longer matches the current product. The assistant isn't lying. It's doing its best with a set of sources that quietly disagree with each other. None of this shows up anywhere the brand can see. No ranking dropped. No traffic dipped. No line on a dashboard so much as twitched. The whole thing — the miss on two engines, the degraded follow-ups, the stale dosage — happened in a room the brand isn't allowed into and doesn't measure. ## Search told you when you were losing. This doesn't. The thing everyone in ecommerce quietly took for granted is that the discovery channel tells you when you're losing. Search does this beautifully: rankings fall, impressions drop, click-through sags, and the numbers arrive before the revenue does. You get a warning. Most of SEO is built on that feedback loop — you can see the wound. The AI channel doesn't give you the wound. It gives you silence. A buyer asks, gets an answer that isn't yours or is a muddled version of yours, and moves on. There's no impression to count, because you were never in the set. You don't lose the sale in a way that registers as a lost sale. You simply never enter a consideration you didn't know was happening. That asymmetry is the heart of the problem. Not that AI is a magical new channel — but that it's the first major discovery surface in twenty years that doesn't come with its own alarm system. The old world handed you a dashboard for losing. This one makes you go looking. Search isn't dead; that's not the claim. There are now two discovery systems running side by side, and most brands are fully instrumented for one and completely blind in the other. The blind one is the one growing. ## Run it yourself Take your own bestseller — the real one, not the one you wish were the bestseller. Ask an assistant what it would recommend for the job that product does. Then ask the follow-up a fussy customer would ask. Read what comes back, across ChatGPT, Perplexity and Gemini. Most founders go quiet halfway through. Not because the answer is a disaster — it rarely is. Because it's the first time they've watched a channel they had no idea they were already competing in. Instrumenting it continuously — engines queried on a schedule, citations tracked, misdescriptions flagged — is what Geoffy's Monitor capability exists for; it's in early access today. But the first look costs nothing but ten minutes and a little comfort. Get the first look done properly: your free GEO Score. ================================================================================ # Four Supplement Brands Own Half the AI Shortlist URL: https://geoffy.ai/blog/four-supplement-brands-own-half-the-ai-shortlist Last updated: 2026-07-01 ================================================================================ Go looking for proper numbers on how brands show up in AI answers — in a category we know well, supplements — and the picture that comes back is blunter than expected. Across roughly 3,800 supplement prompts tested between January and April this year — run against ChatGPT, Claude, Perplexity, Gemini and Google's AI Overviews — four brands took an estimated 47% or more of observed citations between them: Thorne, Seed, AG1 and Momentous. A separate benchmark, narrower in scope, found the same shape inside multivitamins alone: a handful of brands taking the clear majority of mentions, and a substantial minority of the brands tested not appearing at all. Not lower down the list. Not occasionally. Zero. Sit with that. A meaningful slice of a category worth tens of billions is completely absent from the moment a buyer asks the question that matters most: "what should I actually buy?" ## Search was never this brutal The whole logic of SEO was the long tail — a long tail of brands ranking for a long tail of terms. You could be a small player, pick your niche keywords, sit on page two for the broad ones, and still get found. The page was generous: ten blue links, then more below, then more after that. Being twelfth was survivable. AI answers don't have a page two. When someone asks "what's a good magnesium for sleep?", the assistant doesn't return a ranked list of forty options to scroll. It returns a shortlist — three names, maybe five. You are on it, or you don't exist in that conversation. There's the answer, and there's everything the answer left out. That changes the shape of the competition completely. Search rewarded effort spread across a thousand terms. AI discovery rewards being the obvious answer to the question your buyer actually asks. It's winner-take-most, and it resolves much faster than search's slow rank decay ever did. ## What decides the shortlist Here's the part brand teams find hardest to accept: what decides the shortlist isn't who shouts loudest or spends most. Some of the invisible brands in the data outspend the visible ones. What decides it is which brands are legible to the model — consistent claims across their own site and the places the AI trusts, comparison-ready against named alternatives, with public, checkable evidence the model can lean on without taking a risk. The assistant is trying not to be wrong. It reaches for the brands it can stand behind. Most of that work is unglamorous. It's not a campaign. It's making sure that when an assistant checks three sources about your product, it finds the same dose, the same testing claim, the same story every time — instead of three slightly different versions of you it can't reconcile, so it quietly picks someone else. That's the coherence problem in one sentence. ## The silent shift The real shift underneath the numbers: discovery is moving from the search bar to the assistant, without a clear before-and-after moment to mark it. There's no impressions column for an answer you weren't included in. No alert fires. You don't lose the shortlist seat loudly — you never had it, and you find out when your traffic does. The surprising thing is how few brand teams have looked. It takes thirty seconds: open the assistant, ask the question your best customer would ask, read the answer as if you were them. It is reliably the most uncomfortable thirty seconds of the week — and the most useful. You can't fix a gap you've never seen. The brands that notice early get the seat. The rest will spend next year wondering where the demand went. Benchmark data drawn from the 2026 Supplements AI Visibility Index (5WPR), with supporting multivitamin-category benchmarking from Avenue Z. Find out if you're on your category's shortlist: get your free GEO Score. ================================================================================ # Optimising for ChatGPT Is Already the Wrong Goal URL: https://geoffy.ai/blog/optimising-for-chatgpt-is-already-the-wrong-goal Last updated: 2026-06-24 ================================================================================ Almost every ecommerce founder arrives at the same question about AI search. It usually comes out as: "How do I show up in ChatGPT?" The instinct is understandable. ChatGPT was the thing everyone saw first; it became shorthand for the whole shift. If your customers are asking an assistant what to buy, and that assistant is ChatGPT, getting recommended there feels like the whole game. The problem is that the ground has already moved underneath the question. Twelve months ago, ChatGPT accounted for roughly three-quarters of generative AI traffic. The most recent readings put it closer to half, with Gemini's share more than doubling over the same window and Perplexity and Copilot both pulling real volume. None of this means ChatGPT is in trouble — it's still enormous. It means the assumption underneath "optimise for ChatGPT" is quietly expiring. You're no longer trying to be found in one place. You're trying to be found across a handful of them, each reading the web in its own way. ## Why fragmentation changes the job If discovery were consolidating onto a single assistant, the right move would be tactical: learn that one engine's quirks, work them, win the slot. That's the SEO muscle most teams already have — find the algorithm, find the lever, pull it. A comfortable problem, because it's a familiar one. But discovery isn't consolidating. It's splintering. And you can't run four separate optimisation tactics for four engines that each change every few weeks. You'd never finish, and you'd be chasing each one as it moved. The brands that win the recommendation aren't gaming a single engine. They're the ones whose product information is coherent enough that any assistant — reading from any source, on any given week — arrives at the same answer about what they sell and who it's for. ## The real failure mode: illegible, not invisible Here's where it gets uncomfortable. Most catalogues aren't coherent. The specs on the product page say one thing. A marketplace listing says something slightly different. An old blog post or a retailer's feed says a third thing, two years out of date. To a human, that's noise you'd never notice. To an assistant assembling a confident answer, it's contradiction — and a contradictory catalogue is one the model quietly skips in favour of a competitor it can describe without hedging. So the failure mode isn't being invisible to AI. It's being illegible to it. You're in the data. You're just not telling a consistent enough story for any assistant to stake a recommendation on you. That reframing is the whole thing. The work stops being "rank in ChatGPT" — a tactic you re-run forever — and becomes "make my product information agree with itself everywhere it appears" — infrastructure you build once and maintain. And it pays off in every engine at once, including the ones that haven't launched yet. ## The side-by-side test Pick a category you sell in. Ask three different assistants — ChatGPT, Gemini, Perplexity — the same buying question a real customer would ask. Read the three answers side by side. If you get a consistent picture of your brand back, you're in good shape. If you don't appear, or you appear differently in each, that inconsistency is your work for the quarter. It's a slightly deflating exercise the first time. That's rather the point. Discovery is moving from search engines to AI assistants. It's also moving from one assistant to many. Plan for the second shift, not just the first — coherence infrastructure is how, and Geoffy builds it from the catalogue you already have. See how consistently the engines read you: get your free GEO Score — it checks four engines, not one. ================================================================================ # Does ChatGPT Recommend Your Brand? How to Check URL: https://geoffy.ai/blog/does-chatgpt-recommend-your-brand Last updated: 2026-06-18 ================================================================================ The fastest way to find out whether ChatGPT recommends your brand: ask it the questions your customers ask — without naming yourself — and see if you show up. Below is a repeatable method, how to read what you get back, and what to fix when you're invisible. A quick caveat first: AI assistants are probabilistic, and many now fetch the live web for each query. The same question can return different answers on different runs. So this is a pattern-finding exercise, not a single pass/fail — test several phrasings and several runs, and look at the trend. ## How to check, step by step ### 1. Write your buyer-intent prompts List the questions a customer would actually ask an assistant — phrased by need, not by your brand or SKU. For example: - "Best [product category] for [use case] under [budget]" - "What [product] should I buy for [specific situation]?" - "Recommend a [product] for [type of person]" Aim for 10–15 prompts covering your main categories. These are the queries you actually need to win. ### 2. Ask without naming your brand Open a fresh chat (no prior context that might bias the answer) and run each prompt. Don't mention yourself — you want to see who the assistant recommends unprompted. ### 3. Record whether — and how — you appear For each prompt, note one of three outcomes: recommended (named in the shortlist), mentioned (appears but not recommended), or absent. Also note who did get recommended — those are your real AI-discovery competitors. ### 4. Repeat across assistants and runs Run the same prompts in ChatGPT, Gemini, Perplexity, and Claude, and run the key ones two or three times. Coverage varies a lot between assistants, and a single run can mislead. ### 5. Probe the gap Where you're absent, ask a follow-up: "What are the best options from [your brand]?" If the assistant can describe your products accurately, it knows you exist but didn't rank you. If it's vague or wrong, it can't read your data well — a different, more fundamental problem. ## How to read the results - Recommended consistently: you're AI-visible for that intent. Protect it and expand to adjacent queries. - Mentioned but not recommended: the assistant knows you but doesn't have enough signal to prefer you. Usually a structured-data and intent-page problem. - Absent, but accurate on follow-up: discoverable but not competitive — strengthen attributes, use-case coverage, and corroborating sources. - Absent and inaccurate on follow-up: the assistant can't understand your catalogue. This is the foundational fix: crawlability and structured data first. ## Why you're not getting recommended The common causes, roughly in order: 1. Pages aren't reliably crawlable by AI crawlers, so your data never enters the picture. 2. Attributes aren't structured — "premium quality" tells a model nothing it can match against intent. 3. Pages that answer at only one depth — your product pages describe the SKU but never address the situation, constraint or comparison a shopper actually asks about. 4. Stale or contradictory data — wrong prices or availability are strong negative signals and get sources excluded. 5. Thin corroboration — few independent sources describe your products, so the model has low confidence. These map directly onto how AI assistants choose which products to recommend and the groundwork in What is GEO. ## What to do next Fix the foundation in order: make pages crawlable, structure your product attributes and add schema, write the answer to the real intent question onto the product and category pages you already publish, and keep price and availability in sync. Then re-run the same prompt set and watch the pattern move. If you'd rather not do the spot-check by hand every week, this is exactly what an AI-visibility tool automates — running a fixed prompt set across assistants on a schedule and tracking your share of recommendations over time. Geoffy both measures this and fixes the underlying data so the numbers move. ## Frequently asked questions ### How do I check if ChatGPT recommends my brand? Ask ChatGPT the intent-based questions your customers actually use — for example "best magnesium supplement for sleep" — without naming your brand, and see whether your products appear. Repeat across several phrasings, a few assistants, and over time, because answers vary by run. ### Why doesn't ChatGPT recommend my products? Usually because the assistant can't find or can't understand your product data: pages aren't crawlable, attributes aren't structured, the product page only answers at one depth of intent, or price and availability are stale or contradictory. ### Why are the answers different each time I ask? AI assistants are probabilistic and many now fetch the live web per query, so phrasing, timing, and the model version all change the result. Test multiple phrasings and runs and look at the pattern, not a single answer. ### Can I track AI recommendations automatically? Yes. Tools that run a fixed set of buyer-intent prompts across multiple assistants on a schedule turn this manual spot-check into a tracked visibility metric you can watch over time. ## Sources - Aggarwal et al. (2024), GEO: Generative Engine Optimization — https://arxiv.org/abs/2311.09735 - Cloudflare Radar 2025: AI crawler activity — https://blog.cloudflare.com/radar-2025-year-in-review ## About the author Anthony Gale is Co-Founder of Geoffy, a Generative Engine Optimisation platform for ecommerce brands. He has spent more than two decades in ecommerce and digital growth, helping retailers adapt to major shifts in online discovery. ================================================================================ # SEO Rewards Depth. AI Rewards Coherence. URL: https://geoffy.ai/blog/seo-rewards-depth-ai-rewards-coherence Last updated: 2026-06-03 ================================================================================ Spend a fortnight running AI visibility audits across one category — supplements, in this case, a category we know well — and you expect to find patterns. What you don't expect is for the pattern to be this consistent. Here's the headline. The brands that show up in ChatGPT and Perplexity aren't always the biggest. They're not the ones with the best SEO scores, the most backlinks, or the highest organic traffic. They're the ones whose product information is consistent across surfaces. The blog says the same thing as the product page. The reviews use the same vocabulary as the schema markup. The comparison content matches the language a real buyer uses when they ask an AI for a recommendation. That's it. That's the whole pattern. ## The shape of the gap The brands that don't show up — including genuinely well-known names with real customers and real revenue — have rich, beautifully designed sites where every page tells a slightly different story. Different ingredient claims here, different mechanism language there, different dosage framing in the FAQ. The homepage positions the brand one way; the product page another. The blog post written by the in-house nutritionist uses one set of words; the comparison page written by the SEO team uses another. A human reader doesn't notice. A human reader probably appreciates the variety. An AI assistant trying to summarise the brand in roughly two hundred tokens notices. What it sees is contradiction. So it hedges, or it omits — and recommends someone simpler. The most striking part isn't that this happens. It's the size of the gap. Some of these brands have spent years and serious budget on SEO and content — paid teams, agencies, tooling. On every measure that mattered five years ago, they are the content brand in their category. But search rewards depth and breadth; AI search rewards coherence. Those are different optimisation targets, and a site built for the first is not automatically good at the second. Repeatedly, the brand with the deepest content library was being beaten by a smaller competitor whose entire site was a handful of crisp, repeating, machine-readable claims. ## Why the judge changed Classic SEO had a forgiving judge. Google could tolerate internal inconsistency on a domain because its ranking system was built on signals between pages — links, anchor text, query-to-page matches. A clever brand could win those signals without ever resolving its internal contradictions. An AI assistant is a far less forgiving judge. It has to say something coherent about your brand in two sentences, often with no prior context, under a token budget. If the source material disagrees with itself, the assistant doesn't pick one version and commit. It hedges, omits, or picks the brand next door whose material doesn't disagree with itself. The brands performing well in AI answers aren't winning on volume. They're winning on agreement with themselves. There's a kicker in the audit data, too: the brands that did the least SEO theatre — where one person wrote the words and made sure they matched — are often doing best in AI answers. The brands that did the most are often doing worst. Not always. But often enough to be a pattern. Some of those winners almost certainly didn't optimise their way there; they just had a clear voice that propagated. ## The two-minute test Open ChatGPT. Type "best [your category]". Then "best [your category] for [your most distinctive use case]". Then ask it to compare two named SKUs — yours against a direct rival. If your brand isn't in the first three results across those depths, the question isn't whether you have enough content. You almost certainly do. The question is whether your content agrees with itself. Run the same test for your closest competitor while you're there — it tells you who the AI has decided is the canonical voice in your space, and whether you're in the running for the slot. The implication for anyone building now is stark. Treat AI visibility as another optimisation layer on top of existing SEO and you'll spend money without moving much. Treat it as a coherence problem — does my site agree with itself about what I am? — and you can get traction with surprisingly little volume. That's the bet Geoffy is built on, and every audit strengthens it. Want your coherence measured rather than guessed? Get your free GEO Score. ================================================================================ # Shopify Picked agents.md. The Default Is Bland. URL: https://geoffy.ai/blog/shopify-agents-md-the-default-is-bland Last updated: 2026-05-27 ================================================================================ Shopify settled a long-running argument in a single community forum reply. The argument was about which file AI agents should read when they look at a storefront. For most of last year the consensus answer was /llms.txt — the "instructions for the model" file proposed in late 2024. Competing candidates floated alongside it: agents.md, llms-full.txt, various .well-known schemes. None of them mattered enough to win unless a platform with real distribution picked one. Shopify picked one. On production stores, /agents.md now declares itself the canonical agent-facing description of the store, and /llms.txt and /llms-full.txt serve the same content as mirrors of it — /llms.txt says so in its own text. Three files, one source of truth, and the platform has named which one it is. Shopify runs millions of live storefronts. When a platform with that distribution standardises on a filename, the ecosystem follows within a quarter. The protocol war between /llms.txt and /agents.md is now effectively settled on the platform where most consumer commerce happens. If you're building agent-side shopping tools, you write your fetcher against that path. If you're building brand-side agent visibility, the same. Probe well-known Shopify stores and the pattern is consistent: /agents.md serves real content and calls itself canonical, /llms.txt and /llms-full.txt mirror it, and /.well-known/ucp returns 200. Shopify has done the structural work. ## The actually-interesting half of the story Pull the default /agents.md from a few stores and you find the same thing: the template is byte-identical across them. Five sections, fixed wording, and just two merchant variables substituted in — the store's name and its domain. Everything else, including the policy links, is generated from those. The biggest section by word count routes personal shopping agents through Shop Pay. There's a UCP block with discovery endpoints, a read-only browsing section, and a closing block advertising Shopify itself. There is exactly zero product authority. Zero brand voice. Zero curated answers about what this brand sells, what's distinctive about it, or what an agent should say when a shopper asks. A clothing brand, a supplements brand and a homewares brand all ship the same agents.md the day they go live. The platform has made the file canonical. The brand still has to fill it. Most won't — which is precisely the opportunity for the ones that do. ## The twenty-minute fix You override the file through your active theme — check Shopify's current theme docs for the exact template name, since the agent-facing surfaces are still moving. The four things any agents.md needs that the default omits: 1. An identity statement an agent could quote verbatim — "we make X, since Y, distinctive because Z", one paragraph. 2. A curated list of canonical product and category URLs an agent should treat as authoritative. 3. Two or three FAQs with specific answers — the questions shoppers actually ask before buying, answered in the brand's words. 4. A "what to say if asked" block written in your actual voice rather than Shopify's marketing copy. Drafting takes twenty minutes; shipping takes about an hour if your theme is straightforward. It will move the model's output more than another month of paid social. A reasonable test: an agent asked "which is the best X for Y?" should be able to write a paragraph about your brand that you would publish on your own homepage. If it can't, you have an agents.md problem — whether or not you've thought about it in those terms yet. Run curl yourstore.com/agents.md this week. The default is what your customer's assistant is reading right now. And if you'd rather the whole discovery layer — agents.md, structured data, intent-tier content — were engineered and kept in sync automatically, that's what Geoffy does on Shopify. See how agents read your store today: get your free GEO Score. ================================================================================ # Google Just Joined the Agentic Commerce Stack URL: https://geoffy.ai/blog/google-ucp-agentic-commerce-stack Last updated: 2026-05-22 ================================================================================ Google announced the Universal Commerce Protocol from the main stage at NRF, then wrote it up in the slightly bureaucratic register its blog reserves for its own biggest news. Read the post twice and the substance is clear enough — and significant. UCP, co-developed with Shopify and a group of large retailers including Etsy, Wayfair, Target and Walmart, is designed to plug brands directly into AI shopping conversations across Search, Gemini and AI Mode. Three pieces of plumbing stand out. Business Agent lets a shopper chat with a brand on Google Search the way they'd chat with a virtual sales associate — questions, comparisons, recommendations, all inside the conversation. Direct Offers surfaces discounts at the moment of stated intent rather than the moment of bored scrolling. And a new set of Merchant Center attributes feeds the conversational layer with the structured signal it needs to answer confidently. That's Google telling its merchants, in writing, that on its own surfaces the shopping journey happens inside conversations rather than ten blue links. ## The protocols are converging The bigger story is convergence. OpenAI has been wiring checkout into ChatGPT. Shopify has put agent-facing storefronts across ChatGPT, Copilot, Gemini and Google AI Mode. Walmart pulled back from third-party agentic checkout — we covered that separately — but the platform-side trend hasn't slowed; it's accelerated. Now Google is in the same shape, and at pains to position UCP as a protocol the broader ecosystem can implement, not a Google-only standard. The pattern is unmistakable. Whether your stack runs on Shopify, WooCommerce or something bespoke, the AI assistant is now a discovery surface — and increasingly a transaction-adjacent one. ## The tighter question Talk to founders in supplements, beauty, kit and accessories about this news and the reaction follows the same arc in every conversation: first interest, then quiet, then a tighter question. The tighter question is always some version of: how do I show up? The honest answer is uncomfortable for most catalogues, because it doesn't reward the part of the budget they've invested most heavily in. You don't show up in AI answers by having a louder paid social presence. You don't show up by paying more for retargeting. You don't show up because your homepage hero is well-designed. You show up because the assistant — a probabilistic, language-driven model with a recommendation surface — has enough coherent, structured information about your product to summarise it in a paragraph of natural language without hedging. Ingredients explicit. Claims sourced. Comparisons present. The same language across the site, the spec sheet, the FAQ and any third-party listings that carry authority. Schema that doesn't contradict the visible page. Most catalogues haven't done that work. They've done a version of it for Google's old crawler, optimised against a query-and-page-rank model that the assistant has, on the surfaces where shopping now starts, stopped using. ## The window Predictions in this category age badly — the release cadence has outrun everyone's forecasts. But the directional read is hard to escape: within a year, the question for any consumer brand of meaningful size will not be are we doing AI discovery but are we doing it well enough to be the brand the assistant recommends rather than the brand it mentions in passing. The first wave of winners will be the brands that started restructuring their product data this year, not the ones waiting for a board mandate next year. If you haven't opened ChatGPT or Gemini and asked it to recommend a product in your category this week, do it before Friday. The answer is the thing your customers are about to hear. Then get the systematic version: your free GEO Score. ================================================================================ # Three Well-Known Brands. One Obvious Query. Zero Results. URL: https://geoffy.ai/blog/well-known-brands-invisible-in-ai-answers Last updated: 2026-05-19 ================================================================================ A smoke test on the AI discoverability scoring engine we built at Geoffy returned something that looked like a bug: nine of ten brands scanned that morning scored zero on Perplexity. A zero-rate that high demanded a sanity check against the live product. So we picked three of the brands — one footwear, one apparel, one furniture; all real businesses doing seven and eight figures, none fading, none obscure — and queried Perplexity directly with the obvious category-discovery question. No brand name in the query. Just the category and the buyer intent, the same prompt structure our engine sends. Premium men's footwear. Perplexity returned a confident top ten: heritage shoemakers, luxury houses, cult favourites. The test brand — a premium casual name most shoppers in its home market can name without thinking — did not appear. Premium men's lifestyle apparel. Ten brands surfaced, spanning coastal heritage labels and design-led newcomers. The test brand — serious revenue, a flagship store on one of the world's most famous shopping streets, listicle-tier recognition for exactly this aesthetic — did not appear. Premium direct-to-consumer bed frames. Ten names returned. The test brand — arguably the most visible DTC brand in its category over the last five years, the kind of design recognition that ends up in coffee-table magazines — did not appear. Three categories. Three brands you have almost certainly seen advertised. Three zeros. And when we checked the engine's scores against the live Perplexity product, they agreed: the brands really aren't surfaced for these queries. The miss is structural, not a measurement error. ## Why this happens AI products don't have an index the way Google has an index. When you ask Perplexity for "the best premium men's footwear brands in 2026," it isn't ranking pages by keyword relevance. It reads a handful of recent editorial sources, synthesises what they collectively say, and serves you the names that appear most often across that small retrieval set. If your brand isn't named in those sources, you don't appear. It doesn't matter how recognisable you are. It doesn't matter what you spend on Meta. It doesn't matter that the question is one your buyers ask every day. The retrieval system doesn't see you. The brands that lose out aren't the obscure ones. They're the brands whose recognition lives in cultural awareness and paid social rather than in the editorial and structured corpus AI models actually draw on. Years of consumer-awareness building — none of it encoded where the machine looks. And yet the buyer asking that question of ChatGPT or Perplexity or Gemini is exactly the buyer who would have considered these brands if shown the option. Clear product story, real customers, distinctive position — none of it helps if the retrieval doesn't surface it. ## The lesson Visibility in AI commerce is not something you inherit from existing brand strength. It is its own discipline, with its own mechanics — being present, consistently and legibly, at every depth of buyer intent, from the broad question ("I need new slip-ons") through the mid-specific ("waterproof leather slip-ons for travel") to the precise ("premium slip-ons, vegan-leather upper, arch support"). That practice — making a brand the reference answer across those depths — is what we call Reference Engineering, and it's the work Geoffy systematises. If you run a premium brand, run the test yourself. Open Perplexity. Type the question your best customer would ask, without your brand name. See whether you appear. If you don't, you have company — and a problem that won't get smaller as a larger share of buyers start their shopping by asking AI first. Get the structural answer, not just the anecdote: get your free GEO Score. ================================================================================ # AI Found the Product. Walmart Took the Cart Back. URL: https://geoffy.ai/blog/ai-found-the-product-walmart-took-the-cart-back Last updated: 2026-05-15 ================================================================================ Walmart wired around 200,000 SKUs into OpenAI's Instant Checkout. Then their EVP of product shared the data: ChatGPT-driven purchases converting at roughly a third of Walmart's own site rate. He described the result as "unsatisfying" — which is corporate for "we're not doing that again." What Walmart did next is more interesting than the failure itself. They built Sparky — their own assistant — and embedded it inside ChatGPT. The buyer logs into Walmart from inside the chat, carts sync across platforms, and the purchase completes inside Walmart's system. Gemini is next. Owned environment, owned moment of decision. The chat window becomes the shop window — not the till. ## The layers are coming apart Two conclusions follow, and both matter for anyone selling online. First, discovery and checkout are separating — faster than expected. The working assumption across the industry was that assistants would absorb both: the natural endpoint being a checkout inside the answer, with the retailer reduced to a fulfilment partner. Walmart's experiment was the largest single test of that thesis so far, and it didn't hold. Conversion several times worse isn't a tuning problem — it's structural. Buyers need their account, saved payment methods, order history, and returns flow: the entire scaffolding of trust that an owned checkout carries. The chat window doesn't have it, and apparently can't fake it well enough to matter. Second, one number keeps recurring: AI-referred shoppers convert substantially better than traditional-channel visitors — but only once they land on the merchant's own surface. Discovery on the assistant. Decision on your site. That gap is where the value is. ## What this means for merchants The question we hear most, usually from founders running stores between £1M and £20M: "Should I be optimising for the checkout in the chat?" The answer, more confidently than we'd have given it a month before Walmart's data: no. Not yet. Maybe not ever, for most categories. Optimise for being the brand the assistant names. The conversion happens on your turf, where it always has. The naming is the new work — and the naming is where most merchants are presently invisible. Two practical shifts follow: 1. Drop AI-referred conversion rate as your obsession metric. The cohorts are still thin and noisy. The metric that matters is whether the assistant names you at all — visibility, citation presence, mention consistency across the engines that matter. Conversion looks after itself once the visibility is real. 2. Stop treating "agentic commerce" as one thing. There are at least two layers moving independently. The discovery layer is going through real, measurable change right now. The transaction layer is going through the change everyone talks about — and the evidence so far is that it's slower and less certain than the noise implies. The smartest operators have stopped trying to be everywhere in the stack and started picking a layer. The discovery layer — being the answer when a shopper asks — is the one where the change is real today, and it's the one Geoffy is built for. Are the assistants naming you? Get your free GEO Score and find out. ================================================================================ # Most Shopify Stores Are Blocking the Bots They Want to Be Seen By URL: https://geoffy.ai/blog/shopify-stores-blocking-ai-citation-bots Last updated: 2026-05-12 ================================================================================ There's a category of Shopify store right now doing all the visible AI-visibility work — schema, intent-layered content, comparison pages, clean copy — and still getting zero citation hits. Not because the strategy is wrong. Because the plumbing is. ## The one-click own-goal If your store sits behind Cloudflare and you've enabled "Managed robots.txt" — a single toggle in the AI Crawl Control panel — Cloudflare blocks a default list of AI crawlers on your behalf. That sounds reasonable. The problem is that the list lumps together two very different categories of bot. Training crawlers scrape pages to feed back into model training: GPTBot, CCBot, Bytespider, Amazonbot, meta-externalagent. Blocking these is the standard "I don't want my content training someone else's foundation model" posture, and it's defensible. But the same managed list also disallows ClaudeBot — a crawler that does fetch pages to support Claude's answers — alongside the AI-training control tokens Google-Extended and Applebot-Extended. So a toggle most merchants read as "don't train on my content" also removes one of the live fetchers that can cite you. Other citation fetchers, including PerplexityBot and OAI-SearchBot, are not on Cloudflare's managed list at all — you have to check your own robots.txt for those. The practical consequence is the same either way: you can spend months on structured data and answer-shaped content and still be missing from AI answers, because the fetcher gets a 403 when it tries to read the page. Not because anyone made a deliberate choice, but because the toggle was easy and the distinction wasn't. ## The fix: split the posture Decide the training question and the citation question separately, rather than letting one toggle answer both. - Disallow, if you don't want your content training foundation models: GPTBot, CCBot, Bytespider, Amazonbot, meta-externalagent - Reconsider blocking: ClaudeBot, which fetches pages to support answers - Decide deliberately: Google-Extended and Applebot-Extended are training opt-out controls, not citation fetchers. Google states plainly that Google-Extended "does not impact a site's inclusion in Google Search nor is it used as a ranking signal" — so allowing it is a decision about model training, not about visibility - Check separately: PerplexityBot and OAI-SearchBot aren't on the managed list, so your own robots.txt governs them You keep the "don't train on my content" stance — which is what most brands actually mean when they say they don't want AI using their content — without quietly closing the door on the answer surfaces you want traffic from. One currency note: from 15 September 2026 Cloudflare splits crawlers into Search, Agent and Training categories, with search crawlers remaining allowed by default. Re-check your posture after that lands. ## Two related findings worth knowing The Merchant Center → Gemini route is muddier than it looks. The intuitive assumption — "our clean product feed flows to Google Merchant Center, surely that gets us cited in Gemini's chat answers" — doesn't map onto reality cleanly. The feed powers the Shopping graph, which surfaces through search results, Google Shopping and the shopping carousels inside AI Overviews. It's not the same retrieval flow as Gemini's conversational citations. Feed quality lifts your odds in the shopping module; it doesn't necessarily get you mentioned in the prose of a chat reply. Comparison-shape content beats list-shape content for citation. "X vs Y: when to pick each" outperforms "10 best alternatives to X" by a wide margin. The mechanism is question-shape matching: a comparison page gives the model a paragraph already structured as a defensible answer to "should I buy X or Y?", so it can cite cleanly. A list post forces the model to summarise the whole page or pick one item and lose context — and it tends to do neither well. Granularity compounds this: "magnesium glycinate vs citrate for sleep" beats "10 best magnesium supplements", because the narrower comparison maps to a more specific question. ## Why this matters more than it sounds AI search is currently rewarding hygiene work that almost nobody is doing. The flip side is the opportunity: if your competitors have left the Cloudflare default on, you can pass them on visibility for the price of editing one file. That's not a normal SEO situation. Normal SEO doesn't have a setting that hides you from half the answer engines — one that nobody remembers enabling. If you run a Shopify store, open your robots.txt now and check who's allowed. If you're behind Cloudflare, check the AI Crawl Control panel too. The five citation bots above are the ones to allow if you want any chance of appearing in AI answers. For the rest of the discovery stack — llms.txt, structured data, intent-tier content — see how Geoffy prepares Shopify stores for AI discovery. Not sure what's blocked on your store? Get your free GEO Score — crawler access is one of the first things it checks. ================================================================================ # The Ninety-Second AI Visibility Test URL: https://geoffy.ai/blog/the-ninety-second-ai-visibility-test Last updated: 2026-05-06 ================================================================================ There's a test worth running before any conversation about AI visibility, because it takes ninety seconds and it replaces speculation with evidence. Open ChatGPT. Ask: "What's the best [your category] brand?" Then refine: "I'm looking for [specific use case]." Then go specific: "For [exact spec], which brand do you recommend?" That's three depths of buyer intent — Broad, Mid, Ultra. We call them Intent Tiers. Between them, those three queries cover almost every shopping conversation a buyer has with an AI assistant, and they map almost cleanly onto the funnel ecommerce teams already understand. Run it honestly — no brand name in the query — and the results are frequently sobering. Real brands doing seven and eight figures, brands that rank first on Google for their core terms, routinely miss at the Mid tier. Many miss at Broad too. That isn't a corner case. It's the normal state of a market that optimised for one judge for twenty years and is now being evaluated by another. ## Why this happens It isn't that AI dislikes these brands. Assistants work on a coherence model — the sources they pull from, the structured signals they can read, the cross-references they can match. If a brand's product pages don't speak that language, the assistant has nothing to surface. Most ecommerce sites were built for Google. Google rewards keywords, backlinks, page authority. AI rewards machine-readable claims, verifiable specs, and consistent brand identity across the web. Different game, different ground. The result: a brand that ranks first on Google for "cycling bib shorts" can be invisible when someone asks ChatGPT for cycling bib shorts. That's not the brand being punished. That's the brand showing up in the wrong room. ## Reading your result The tier where you miss tells you which conversation to have internally: - Miss at Broad — a structural problem. The assistant can't establish who you are or that you belong in the category at all. This usually means missing or contradictory foundational data across your site and the sources AI reads. - Miss at Mid — a content problem. You exist, but nothing connects your products to the use cases buyers actually describe. Comparison and use-case content, structured properly, is the lever. - Miss at Ultra — a specification problem. Your product data doesn't carry the precise, machine-readable attributes that let an assistant match you to an exact requirement. Three different conversations, all starting from the same three queries. ## Why acting early compounds The next twelve months will separate brands that worked this out early from ones that didn't. Not because AI is replacing search wholesale — at the high-recall end of the funnel, brand strength still matters — but because brands that show up consistently across Google, ChatGPT, Perplexity, Gemini and the agentic layer behind them compound in one direction, and brands missing from two of the five compound in the other. Most ecommerce leaders know this directionally. What we see in conversations is hesitation about where to start, and pessimism about whether it's worth investing before the gap shows up on a dashboard. The trouble is that this channel doesn't have a dashboard — the loss is silent. The ninety-second test is the cheapest instrumentation you'll ever install. Take the ninety seconds today. The result tells you which conversation you should be having — and how soon. When you're ready to go deeper than three queries, Geoffy's approach to structured product data picks up exactly where the test leaves off. Prefer a full diagnostic across engines and intent tiers? Get your free GEO Score. ================================================================================ # AI Is a Discovery Channel, Not a Checkout Channel URL: https://geoffy.ai/blog/ai-is-a-discovery-channel-not-a-checkout-channel Last updated: 2026-05-01 ================================================================================ For the past eighteen months, anyone arguing that AI was eating ecommerce search had to fight for it. The SEO industry is large, conservative and well-paid, and most of it spent the year telling clients to keep doing what they were doing — publish content, build links, optimise for Google, ride out the LLMs. That argument is now effectively over — and it was the SEO industry's own paper of record that ended it. In April, Search Engine Land covered a survey of just over 1,000 US consumers run by Exploding Topics. The headline numbers: - 77% have used AI to shop in the past six months. - 44% start their purchase journey inside an AI assistant — not Google, not Amazon. - Another 44% start elsewhere and bring AI in before they buy. - Only 2% actually check out through the AI. - 31% say they wouldn't authorise an AI agent to spend a single dollar on their behalf. Most of the coverage read those numbers as agentic commerce is here. That's the headline-friendly read, and it isn't quite right. ## The useful read: discovery, not checkout The far more useful read is this: AI is now a discovery channel, not a checkout channel. Roughly half of shopping journeys begin in an assistant. Almost none of them end there. That distinction matters because much of the AI-commerce industry has been building for the wrong half of it. The high-profile bets — instant checkout inside chat, agent-payment protocols, wallet integrations — are all checkout bets, moving the transaction inside the AI surface. The consumer data says shoppers haven't agreed to that yet, and early launches keep running into the wall of distrust that the 31% figure represents. Meanwhile the thing consumers are doing — asking an assistant "what's the best X for Y?" — has already become the way half of America starts shopping. And most ecommerce teams aren't optimised for it at all. The line from the Search Engine Land piece worth pinning above your desk: "a huge base of potential customers are using AI as a starting point, so it is imperative that your brand gets organically mentioned." The publication that built the SEO playbook, naming the new game. ## Two things this means in practice First: the surface area is wider than most teams think. ChatGPT leads AI shopping usage, but Gemini, Perplexity, Copilot and Claude each carry a meaningful share of shoppers. If you're auditing visibility in one engine only — and most agencies audit in one, usually ChatGPT — you're flying blind on a large share of the surface. The work has to be cross-engine or it isn't really being done. Second: the mechanism is different from ranking. SEO operates against a ranked list — page one, position three, climb the SERP. AI search operates against a single recommendation. The model picks one or two products, names them, and moves on. There is no page two. The discipline that gets you into that slot is closer to engineering than to content marketing: product information clean, structured, and comparable enough that the model can pick you accurately. We call that practice Reference Engineering; the industry calls the broader category GEO. Same problem, different vocabulary. ## The fifteen-minute exercise If you sell online, the most useful thing you can do this week is stop debating whether this matters and check whether your product information is actually answerable. Run the questions your customers ask before they buy through ChatGPT, Gemini and Perplexity. See whether you turn up. See whether you turn up accurately. That fifteen-minute exercise will tell you more than any agency proposal you'll receive this quarter. The argument is over. The work has only just started. Source: Search Engine Land, "New data: 77% use AI to shop. Nearly 1 in 3 won't let it spend," April 2026. See how you show up across the engines. Get your free GEO Score. ================================================================================ # The Citation Layer Is Getting Smaller URL: https://geoffy.ai/blog/the-citation-layer-is-getting-smaller Last updated: 2026-04-24 ================================================================================ A small shift in a model's behaviour is rarely front-page news. But every so often one lands that you have to sit with before the implications catch up with you. Recent ChatGPT releases have been citing measurably fewer domains per answer than they did before — a meaningful slice of the citation layer gone within a release cycle — while at the same time running more internal search queries per response, not fewer. The pattern is clear: the model is looking in more places and linking to fewer of them. ## Searching harder, filtering harder The instinct is to read fewer citations as retreat — the model doing less work. The fan-out numbers say otherwise. More queries plus fewer citations means the model isn't doing less; it's filtering more aggressively. It examines more sources and keeps fewer. Anyone who has watched search since 2008 will recognise the shape. Every major Google update did some version of the same thing: crawl more, rank fewer, keep perceived quality high by cutting the long tail. The difference is that Google's cuts fell mostly on spam and affiliate farms — you could see what was dropped, and nobody missed it. With AI answers, the criteria for "worth citing" are less visible, the pool of potentially citable content is far broader, and the narrowing happens in the dark. ## What "authority" means when the model decides The research coming out of GEO studies this year is converging on a consistent picture of what language models prefer to cite. Four signals keep appearing: strong domain authority, structured and unambiguous factual content on the page itself, broad consensus across independent sources, and — most interestingly — original data or analysis the model can attribute to you specifically. What the research is not finding is weight behind the things many brands have been paying for: mentions in listicles, guest posts, inclusion in "top 10" articles on second-tier blogs. The citation layer doesn't appear to care much about those. And as the citation pool tightens, it cares less. The implication is uncomfortable. The old playbook — build backlinks, place mentions, get covered anywhere — is being replaced by something closer to: be a primary source, or be invisible. You don't earn a citation by being talked about. You earn it by being worth reading directly. ## The narrowing is already visible in shopping answers Ask an assistant for a product recommendation and you get three names. Not twenty. Not page one of Google. Three brands, ranked and reasoned, with a clear preference — and the direction of travel is towards fewer, not more. If you're one of the brands named, this is extraordinary news: the model has pre-qualified the shopper before they reach you, and the less the model cites, the higher the intent of the traffic you receive. If you're a brand that used to live in the long tail — reachable through a comparison article, a forum thread, an outbound link from a review site — you are being quietly excluded from conversations you used to at least be part of. Not by anyone deciding you don't deserve it. By a filter that gets more ruthless with each release. ## What to actually do Three practical moves follow from this. First, stop over-investing in the signals models weight least. Paid mentions on mid-tier blogs and keyword-density exercises were already flattening; a narrowing citation layer flattens them further. Second, become a primary source worth citing. Own data about your category, publish it, keep it current, and make it structured and unambiguous. The brand most likely to be cited on "best winter gloves for cycling commuters" is the one that measured thermal performance and published the results — not the one with the most affiliate placements. Third, plan for scarcity. Assume the assistant names one, two, maybe three brands in your category, and work backwards: what would you have to be, and be known for, to hold one of those slots? Making your product information coherent and machine-readable is the foundation everything else stands on. For the whole of the SEO era, the web's economic model assumed abundance — page one, then page two, then a long tail that gave smaller brands a way in. The AI era is moving towards scarcity. There is no list. There is an answer, and you are in it or you aren't. Find out whether you're in the answer. Get your free GEO Score. ================================================================================ # AI Doesn't Recommend the Best Products. It Recommends the Ones It Can Understand. URL: https://geoffy.ai/blog/ai-recommends-what-it-can-understand Last updated: 2026-04-16 ================================================================================ Ask an AI assistant a simple buying question — "what's a good magnesium supplement for energy and sleep?" — and you'll get a confident answer. Three brands, clear reasoning, even a breakdown of the different types of magnesium and what each is good for. It's a better experience than scrolling to page two of Google, which is what product research used to be. Then do the thing most people never do: check who made the list, and why. Run that test across categories and a pattern emerges quickly. Brands with well-structured product data — proper specs, comparison information, clear attributes a machine can parse — appear. Brands whose catalogues are beautiful but unstructured barely scrape in, however large their marketing budget. And dozens of genuinely good brands in every category appear nowhere at all. Not ranked low. Absent. ## The uncomfortable implication The AI isn't recommending the best products. It's recommending the products it can understand. That distinction matters more than most ecommerce teams realise. Twenty years of optimisation effort has gone into Google — backlinks, domain authority, keyword coverage — and it worked. But AI assistants don't read the web the way a search crawler ranks it. A language model doesn't care about your PageRank. It cares about whether it can parse your product information well enough to stake a recommendation on it. The pattern repeats in category after category. Wetsuits: some brands appear with detailed reasoning about neoprene thickness, seam construction and thermal lining, while brands with excellent products are simply missing — not because the products are worse, but because the assistant couldn't find enough structured information to recommend them with confidence. Beauty: brands with properly structured ingredient lists, skin-type matching and clear attributes show up; brands hiding everything behind lifestyle imagery and "discover your glow" messaging are invisible. Cycling kit, electronics accessories, home goods — the same gap, over and over. ## It's not about brand size. It's about data structure. This is the part that should get your attention. Some of the brands assistants recommend are relatively small — they just built their catalogues properly. Some of the brands assistants ignore are household names spending millions on advertising. The ad spend is irrelevant to the model. The Instagram following is irrelevant. What matters is whether an AI system can look at your product data and understand what you sell, who it's for, and why someone should buy it. And the shift is happening at scale. Traditional search volume is falling as AI referrals to retail sites grow at extraordinary rates, and shoppers increasingly start purchase research inside an assistant rather than a search box. People haven't stopped shopping — they've stopped starting every purchase on Google. The economics of visibility change with it. On a Google results page you had ten organic slots to aim for. In an AI answer you have perhaps two or three mentions. Possibly zero. The competition for visibility hasn't just moved — it has concentrated. ## What to do about it Most brands have no idea this gap exists, because they're still measuring success by Google rankings and return on ad spend. Meanwhile a growing share of their potential customers are asking assistants what to buy and getting answers that don't include them. The starting point costs nothing: ask ChatGPT or Perplexity to recommend products in your category, without your brand name in the query. If you're not in the answer, the fix isn't more content or more ads — it's making your product information structured, specific and consistent enough for a machine to understand. That's the discipline GEO exists for, and it's what Geoffy does for Shopify and WooCommerce stores. The better product should win. Right now, the better data does. Close that gap before your competitors do. Want to see where you stand today? Get your free GEO Score. ================================================================================ # GEO for WordPress: What WooCommerce Merchants Need to Know URL: https://geoffy.ai/blog/geo-for-wordpress-what-woocommerce-merchants-need-to-know Last updated: 2026-03-12 ================================================================================ import { appGetStartedHref } from "../../data/site"; WordPress powers a significant share of ecommerce. Here's how AI discovery applies. WordPress and WooCommerce run a significant share of the world's online stores. And like their Shopify counterparts, most of them are completely invisible to AI assistants. That's not a criticism — it's just the current state. The infrastructure AI systems need to recommend products confidently doesn't exist out of the box on any ecommerce platform. It has to be built. ## What's different about WordPress for AI discovery WordPress stores tend to have one advantage Shopify stores often lack: more content flexibility. It's typically easier to publish discovery-oriented pages, create organised content structures, and build out a first-party layer without fighting the platform. The challenge is that most WordPress ecommerce setups are still built around SEO assumptions. Products are optimised for keywords, categories are organised for browsing, and the content that exists is written for human readers rather than machine reasoning. That's fine for search. For AI discovery, it's not enough. ## The gap that needs filling AI assistants answering shopping questions need to find organised, intent-aligned context around your products — not just individual product pages. When a customer asks Perplexity "what's the best standing desk for a small home office under £500?", the assistant needs to be able to evaluate your relevant products against that specific question. If that context doesn't exist on your domain, your products simply aren't in the conversation. ## What AI-ready looks like for WordPress The core requirement is a first-party discovery layer: structured pages on your own domain, organised around the questions your customers are actually asking AI assistants, and kept current as your catalogue changes. This doesn't require rebuilding your WordPress site or changing your theme. It's a layer that sits alongside your existing store, adds new discovery surfaces on your domain, and keeps them aligned with your product catalogue over time. ## The early mover opportunity AI discovery is still early for WordPress merchants. That's an advantage, not a reason to wait. The stores building AI discovery infrastructure now are capturing visibility in a channel that's growing fast — and the early structural advantage compounds over time. Start with WordPress → ## Frequently asked questions ### Does this work with WooCommerce specifically? Yes. Geoffy supports both WordPress and WooCommerce stores. See WordPress solutions → ### Do I need a developer to set this up? No. Geoffy connects to your existing WordPress catalogue and engineers the product pages you already publish, without requiring development work or theme changes. ### Will it affect my existing WordPress SEO? GEO adds to your existing SEO setup — it doesn't interfere with it. Because the work lands on the pages you already rank, deeper and better-structured content tends to help organic visibility too. ### How is WordPress support different from Shopify? The underlying GEO infrastructure is the same. The integration connects directly to your WordPress/WooCommerce catalogue. ## About the author The Geoffy team works with ecommerce agencies and merchants to build first-party AI discovery infrastructure. geoffy.ai ================================================================================ # How Agencies Can Offer GEO as a Service URL: https://geoffy.ai/blog/how-agencies-can-offer-geo-as-a-service Last updated: 2026-03-12 ================================================================================ AI discovery is the next major channel. Here's why it belongs in your agency offer — and how to position it. Every few years, a new channel emerges and agencies that move early build durable revenue around it. SEO in the early 2000s. Paid social in the 2010s. Each time, the agencies that established expertise early won long-term retainer relationships. The ones that waited became order-takers. AI product discovery is that channel right now. ## Why GEO is a natural agency service Your clients are already asking about AI. They've seen the coverage, they've used ChatGPT themselves, and they're wondering why their products aren't showing up when customers ask for recommendations. They don't know what to do about it. That's your opening. GEO — Generative Engine Optimisation — is the practice of structuring product content so AI assistants can understand and recommend it. It sits alongside SEO, not in competition with it. And unlike paid channels, it builds compounding organic visibility over time. For agencies, it's a high-value, defensible service that's genuinely hard to replicate in-house quickly. Your clients can't easily DIY it. And once you've built their AI discovery infrastructure, you own the ongoing management relationship. ## What a GEO service looks like in practice The core deliverable is a first-party discovery layer built on the client's own domain — structured pages organised around the questions their customers ask AI assistants, maintained as their catalogue evolves. Done well, this service includes an initial discovery audit (what's the current AI visibility picture?), a structured build of intent coverage (how many buying questions does the catalogue need to answer?), an ongoing management retainer (as products change, that content needs to stay current), and regular reporting on AI referral trends and coverage growth. It's not a one-time project. It's an ongoing relationship — which is exactly what strong agency retainers are built on. ## How to position it to clients Don't lead with the technology. Lead with the business problem: your products aren't showing up when customers ask AI assistants for recommendations. Your competitors might be. The conversation shifts when clients see that AI-referred traffic already converts at higher rates than most channels — and that the brands building AI discovery infrastructure now are compounding an early advantage that will be hard to close later. ## Why Geoffy works for agencies Geoffy is built with agency workflows in mind. You can manage multiple client catalogues from a single workspace, keep the content and structured data on each client's own product pages current, and report on intent coverage over time. You don't need to build the infrastructure yourself. Geoffy provides it — you provide the strategy, the account management, and the client relationship. See the agency offer → ## Frequently asked questions ### Is GEO a replacement for SEO services? No. GEO complements SEO and sits alongside it. Agencies can offer both — they address different parts of the discovery landscape. ### Is this suitable for smaller agency clients? GEO scales from single-store merchants up to large multi-product catalogues. The Starter plan is accessible for smaller clients; Scale plans suit larger accounts. ### What does ongoing management involve? Keeping the client's product-page content and structured data current as the catalogue changes — new products, updated specs, pricing changes, discontinued lines. It's the retainer component of the service. ### How does Geoffy support agency billing and reporting? Geoffy's agency plan includes multi-client workspace management and coverage reporting. Talk to us about agency access → ## About the author The Geoffy team works with ecommerce agencies and merchants to build first-party AI discovery infrastructure. geoffy.ai ================================================================================ # How AI Assistants Choose Products to Recommend URL: https://geoffy.ai/blog/how-ai-assistants-choose-products-to-recommend Last updated: 2026-03-10 ================================================================================ What's actually happening when an assistant recommends one product over another. When a customer asks ChatGPT "what's the best espresso machine for a beginner?" something very different from a Google search is happening. There's no list of ranked links. The assistant analyses, weighs options, and produces a shortlist. One brand makes it. Most don't. For ecommerce teams, the question is obvious: what decides who gets recommended? ## The competitive surface just got smaller Traditional search returns ten blue links. AI assistants typically surface two to five products per query. Missing that shortlist doesn't mean ranking lower — it means not being considered at all. The stakes of product visibility have changed fundamentally. ## What AI systems need to make a recommendation AI assistants are not matching keywords. They are reasoning about fit. When a customer asks about running shoes for flat feet, the assistant is trying to answer: does this product actually address this need? Can I say that with confidence? Products that are easy to reason about get recommended. Products that are vague, inconsistent, or poorly structured get skipped — not because they're worse, but because the assistant can't evaluate them confidently. ## The clarity gap Most product pages were written for human browsers, not machine reasoning. A description like "premium quality, perfect for everyday use" gives an AI assistant almost nothing to work with. It can't determine use case fit, compare it against alternatives, or include it in a specific recommendation with confidence. The brands winning in AI discovery are the ones whose products are specific, structured, and consistent — making it easy for AI systems to say "yes, this fits" without uncertainty. ## Why clusters matter AI assistants rarely recommend one product in isolation. They build shortlists — two to five options that fit the query from different angles. To be included, your products need to be part of a well-defined discovery space, not floating in isolation. That's why the unit of AI visibility isn't the product page. It's the discovery layer that organises your products around the questions customers actually ask. ## What this means for your store The brands showing up in AI recommendations today didn't get there by accident. They have clear product data, well-organised discovery coverage, and content that makes it easy for an assistant to say "this fits." That infrastructure can be built. But it takes more than SEO. Geoffy helps ecommerce teams build the discovery layer that gets products into AI recommendations. See how it works → ## Frequently asked questions ### How do AI assistants choose which products to recommend? They reason about fit — evaluating how well a product matches the user's specific question based on the clarity and completeness of available product information. ### Why do some products get recommended and others don't? Products that are easy for AI systems to interpret and evaluate confidently are more likely to appear in recommendations. Vague or inconsistently structured products give assistants less to work with. ### Is this the same as SEO? No. SEO is about ranking pages. AI discovery is about whether assistants can understand and recommend your products within generated answers. ### What is GEO's role here? Generative Engine Optimisation (GEO) is the practice of structuring product content so AI assistants can interpret and recommend it confidently. Learn more about GEO → ## About the author Anthony Gale is Co-Founder of Geoffy, a Generative Engine Optimisation platform for ecommerce brands. He has spent more than two decades working in ecommerce and digital growth, helping retailers adapt to major shifts in online discovery. ================================================================================ # How to Prepare Your Shopify Store for AI Discovery URL: https://geoffy.ai/blog/how-to-prepare-your-shopify-store-for-ai-discovery Last updated: 2026-03-10 ================================================================================ import { appGetStartedHref } from "../../data/site"; The window to build early AI visibility is open now. Here's where most Shopify stores stand. AI assistants are already recommending products to your customers. ChatGPT, Perplexity, Gemini, and others are answering shopping questions every day — and producing shortlists that include some Shopify stores and exclude most. The difference between the stores getting recommended and the ones being ignored isn't product quality. It's discovery readiness. ## Where most Shopify stores are right now Shopify is an excellent platform, but it was built for a search-led world. Most Shopify stores have strong product pages, solid SEO basics, and well-managed catalogues. What they don't have is a discovery layer that AI assistants can actually use. The issue isn't what's on the product page. It's what's missing around it. AI assistants don't just read individual product pages — they look for organised, intent-aligned context that helps them answer specific questions confidently. Most Shopify stores have no such layer. ## What AI-ready actually means Being AI-ready for a Shopify store means three things: Your product data is structured so machines can reason about it, not just humans. Your catalogue is organised around the questions customers ask AI assistants, not just product categories. And that structure lives on your own domain, as first-party content AI systems can trust and reference. Stores that have this in place are appearing in recommendations. Stores that don't are invisible to the fastest-growing discovery channel in ecommerce. ## Why now matters AI-referred ecommerce traffic grew dramatically through 2024 and 2025, and the quality of that traffic is exceptional — longer sessions, higher conversion rates, and more revenue per visit than almost any other channel. The brands building AI discovery infrastructure today are compounding an early advantage. Waiting for the channel to mature before acting is the equivalent of waiting until 2010 to start SEO. ## The good news for Shopify merchants You don't need to rebuild your store. You don't need to change your theme or restructure your product pages. The discovery layer sits alongside your existing store — published on your domain, aligned with your products, and maintained as your catalogue evolves. Geoffy connects directly to your Shopify catalogue and builds that layer for you, on the product pages you already have. The first optimised pages can be live in minutes. Start with Shopify → ## Frequently asked questions ### Does preparing for AI discovery mean rebuilding my Shopify store? No. The discovery layer sits alongside your existing store without requiring theme changes or a rebuild. ### Will this affect my existing SEO? GEO complements SEO — it doesn't interfere with it. The same structured, first-party content that helps AI assistants also supports organic search. ### How quickly can I get started? Geoffy can connect to your Shopify catalogue and publish the first optimised product pages in minutes after setup. ### Is it worth acting now or waiting until AI discovery is bigger? The brands building AI discovery infrastructure now are compounding an early advantage. The channel is growing fast and the early movers are already ahead. ================================================================================ # What is Generative Engine Optimisation (GEO)? URL: https://geoffy.ai/blog/what-is-generative-engine-optimisation-geo Last updated: 2026-03-10 ================================================================================ The next evolution of ecommerce visibility in the AI era. For more than two decades, ecommerce visibility has been dominated by search engines. If a brand wanted customers to find its products online, it invested in Search Engine Optimisation (SEO): optimising pages to rank for keywords in search results. But discovery behaviour is changing rapidly. Consumers are increasingly turning to AI assistants such as ChatGPT, Gemini, and Perplexity to ask questions directly: - What's the best espresso machine for beginners? - Which standing desk should I buy for a small office? - What are the best lightweight backpacks for travel? Instead of showing a list of links, these systems generate answers and recommendations. Data already shows how quickly this behaviour is emerging. AI-referred retail traffic grew 35x between July 2024 and May 2025, according to Adobe Analytics research on ecommerce referral patterns. This shift is creating a new optimisation discipline: Generative Engine Optimisation (GEO). !A person using an AI assistant on a laptop to research products ## The shift from search engines to answer engines Traditional search engines return ranked lists of pages. Users must evaluate those pages themselves. The classic SEO-driven journey looks like this: Keyword -> Search engine -> List of links -> User comparison -> Purchase AI assistants change this dynamic. Instead of presenting links, they analyse information and generate a direct recommendation. The discovery journey increasingly looks like this: Question -> AI analysis -> Synthesised answer -> Product recommendation -> Click -> Purchase In this model, the AI assistant becomes a decision layer between consumers and brands. And the competitive surface is much smaller. AI systems typically surface only a small shortlist of products per query. If a product does not appear in that shortlist, it may never be considered by the customer at all. The key question for ecommerce brands is therefore no longer simply: Can customers find my pages? Instead it becomes: Can AI systems understand and recommend my products? !Traditional search results compared to an AI assistant answer ## What is Generative Engine Optimisation? Generative Engine Optimisation (GEO) is the practice of structuring product information so that AI systems can interpret, evaluate, and recommend it. Where SEO focuses on ranking pages, GEO focuses on making products intelligible to AI systems. The term gained wider attention following a 2024 academic study by researchers from Princeton University, Georgia Tech, and IIT Delhi. Their research showed that structured, clearly attributed content increased visibility in generative AI responses by up to 40% compared with unstructured alternatives. At a practical level, GEO involves three core layers. ### 1. Discovery signals AI systems still rely on technical signals to locate and interpret content. Important discovery infrastructure includes: - XML sitemaps - internal linking - structured metadata - emerging standards such as llms.txt These signals help AI crawlers identify what content exists on a site. Cloudflare Radar research shows that AI-driven crawling activity increased more than 15x in 2025, as AI assistants increasingly fetch live information to answer user questions. If a page cannot be discovered by these crawlers, it cannot be recommended. ### 2. Semantic understanding Large language models interpret content differently from traditional search engines. Instead of focusing primarily on keyword matching, they attempt to understand: - product attributes - relationships between variants - category structures - suitability for specific use cases Structured formats such as JSON-LD schema and product attribute data help AI systems interpret these relationships. For example, a product description that says "premium quality running shoe" provides little machine-readable context. But structured attributes such as: - cushioning level - stability type - terrain compatibility - weight - price range allow an AI system to reason about whether the product fits a particular query. !Structured product data attributes mapped into a knowledge graph ### 3. Intent alignment Consumers rarely ask AI assistants for specific SKUs. Instead, they ask intent-driven questions such as: - Best travel backpacks for carry-on luggage - Running shoes for flat feet - Standing desks under £500 To answer these questions effectively, AI systems prefer content organised around clusters of products that match the intent. This means discovery increasingly depends on intent-aligned product groupings, rather than isolated product pages. ## Why traditional SEO is no longer enough SEO remains important, and many principles of good SEO (clear structure, relevant content, technical accessibility) also support GEO. But search engines and AI assistants operate differently. Search engines: - rank pages - return links - let the user compare options AI assistants: - interpret product attributes - synthesise information - generate recommendations This difference changes the optimisation problem. A page that ranks well in traditional search results may still fail to appear in AI recommendations if the underlying product data lacks clarity or structure. ## How AI assistants evaluate ecommerce content Large language models build internal representations of information, often described as knowledge graphs. These graphs map relationships between entities such as: - products - brands - categories - attributes - use cases A well-structured product might be represented conceptually like this: ``text Running Shoe ├ Cushioning -> Maximum ├ Stability -> High ├ Arch Support -> Structured ├ Terrain -> Road ├ Weight -> 280g └ Price -> £120 `` When a user asks about running shoes for flat feet, the AI system can directly evaluate which products have attributes that match the requirement. If those attributes are missing or poorly structured, the AI may not have enough signal to recommend the product. !Abstract representation of an AI knowledge graph connecting product entities ## GEO vs SEO The relationship between SEO and GEO is best understood as an evolution rather than a replacement. SEO GEO Focus Keywords User intent Output Ranked pages Product recommendations Competition Pages compete for ranking Products compete for recommendation Results List of links Synthesised answer Decision maker The user The AI system Optimisation unit Page Product data Both approaches will coexist. However, as AI assistants become a larger part of product discovery, GEO will become an increasingly important layer of ecommerce visibility. ## Why this matters for ecommerce brands now AI-driven discovery is still early, but several signals suggest it is growing quickly. - AI-referred ecommerce traffic has increased dramatically in recent years. - AI crawlers are becoming significantly more active across the web. - Structured product data improves the likelihood of appearing in AI responses. For merchants, the implication is straightforward: Products that are easier for AI systems to understand are more likely to be recommended. ## The next era of ecommerce visibility The history of ecommerce discovery has been shaped by technological shifts. First came web directories. Then search engines transformed how consumers found products. Now AI assistants are becoming the next major interface between consumers and the internet. In this new environment, success will depend not only on ranking pages but on ensuring products can be understood, evaluated, and recommended by AI systems. Generative Engine Optimisation is the discipline that makes this possible. ## Frequently asked questions ### What is Generative Engine Optimisation (GEO)? GEO is the practice of structuring product information so AI assistants such as ChatGPT, Gemini, and Perplexity can understand and recommend products. ### Is GEO the same as SEO? No. SEO focuses on ranking pages in search results. GEO focuses on making products understandable to AI systems so they can be recommended in answers. ### Does GEO replace SEO? No. Both will coexist. However, AI-mediated discovery is growing rapidly, which means product visibility increasingly depends on AI understanding. ### What platforms does GEO apply to? GEO principles apply to any ecommerce platform. Shopify and WordPress are currently the most common environments where structured GEO optimisation is being adopted. ### How do I get started with GEO? Start with a product data audit: - Are product attributes structured clearly? - Are relationships between products defined? - Do your product and category pages answer the common intent queries? Platforms such as Geoffy help automate this process. ## Sources - Adobe Analytics: AI-driven retail traffic insights https://business.adobe.com - Aggarwal et al. (2024), GEO: Generative Engine Optimization https://arxiv.org/abs/2311.09735 - Cloudflare Radar 2025: AI crawler activity https://blog.cloudflare.com/radar-2025-year-in-review ## About the author Anthony Gale is Co-Founder of Geoffy, a Generative Engine Optimisation platform for ecommerce brands. He has spent more than two decades working in ecommerce and digital growth, helping retailers adapt to major shifts in online discovery. ================================================================================ # Why Product Pages Alone Won’t Work for AI Search URL: https://geoffy.ai/blog/why-product-pages-alone-wont-work-for-ai-search Last updated: 2026-03-10 ================================================================================ Most ecommerce sites organise their catalogue around one page type: the product page. That structure worked in a search-led world. It breaks in an AI-led discovery world. !A shopper researching products on a laptop ## The mismatch between questions and product pages Product pages answer one question well: Tell me about this exact product. AI assistants are usually asked a different question: What products are best for this situation? For example, when a user asks, What’s the best espresso machine for beginners?, the assistant needs to: - compare multiple products - evaluate ease of use - assess budget fit - present a shortlist A single product page cannot do this alone. ## How AI assistants build recommendations A typical recommendation flow is: - interpret user intent - identify relevant category context - evaluate multiple products against intent - generate a shortlist So AI systems tend to assemble product clusters, not single-page answers. ## Why traditional ecommerce architecture struggles Most stores are built around: - product pages - broad category pages Category pages often list all items in a class, but do not align to specific intent queries such as: - best running shoes for flat feet - standing desks under £500 - lightweight backpacks for travel The AI then has to reconstruct an answer from pages that were not built for that question. ## The tempting fix, and why it fails The obvious response is to build a second set of pages — one per buying question, sitting alongside the catalogue. Best running shoes for flat feet. Standing desks under £500. Pages shaped like the answer. It reads as though it should work. It does not, for three reasons. You now maintain two descriptions of the same product, and they start disagreeing the day the catalogue changes. The new pages carry none of the links, reviews or purchase history that make the real product URL trustworthy, so you are asking a model to cite your weakest page instead of your strongest. And a page built for machines that no human has reason to visit is a doorway — an old pattern with a long record of being discounted once platforms notice it. The diagnosis above is right. The prescription is wrong. ## What actually closes the gap The mismatch is not that you have too few pages. It is that the pages you already publish only answer at one depth. A product page can carry the broad question and the ultra-specific one at the same time: an overview that names the situation the product is for, structured attributes a model can compare against, honest pricing and availability, and the reasoning a shopper with a particular constraint actually needs. Those are blocks on the page, not destinations of their own. The category page does the group-level version of the same job. You already publish it, it already ranks, and it is already the natural landing point for "what’s the best X for Y?" — provided it explains why these products belong together and what separates them, rather than listing everything in the class. Do that and the assistant no longer has to reconstruct an answer from pages that were not built for the question. The answer is on the page that was already going to win. ## Why coherence beats volume Assistants are not rewarding page count. They reward pages they can quote without hedging: - relevant products grouped with a stated reason - an explanation of which situation each product suits - structured comparison context that matches the visible text The page should resemble the answer the assistant is trying to give — on the URL that already carries your authority. ## The role of GEO Generative Engine Optimisation (GEO) is the process of structuring content so AI assistants can: - understand product attributes - evaluate relevance to real questions - recommend products within answers As AI-mediated discovery grows, the competitive requirement is depth and coherence on the pages you already own — not a bigger site. ## Frequently asked questions ### Are product pages still important? Yes. They remain core for detail and conversion, but they do not cover the full AI discovery layer. ### Should we build a separate page for every buying question? No. It splits your catalogue into two versions that drift apart, and it moves content off the URL that carries your links, reviews and trust signals. Answer the buying question on the product and category pages you already publish. ### Why do AI assistants favour some pages over others? Because those pages mirror the question and provide comparison context the model can check — grouped products, a stated reason they belong together, and structured data that agrees with the visible text. ### What does that look like in practice? An overview that names the situation the product is for, structured attributes, honest pricing and availability, and guidance for the shopper with a specific constraint — all as blocks on the canonical page rather than pages of their own. ## About Geoffy Geoffy engineers the canonical product page so AI assistants can read, trust and cite it — answer-ready content across the full range of buyer intent, with structured data that matches the visible page. One page, not a parallel site of AI pages. ================================================================================ # The Provenance Report: Why AI Discovery Content Fails Before It Is Written URL: https://geoffy.ai/whitepapers/the-provenance-report Last updated: 2026-08-19 ================================================================================ ## Executive summary There is a widely held assumption about AI product discovery that the largest available dataset contradicts. The assumption is that AI assistants prefer community sources — Reddit threads, review platforms, forums — over brand-owned pages, and that brands are therefore fighting for scraps in their own category. That is not what the data shows. Roughly 57% of AI citations go to company-operated web properties. Brand-owned content is the majority of what gets cited. This should be encouraging, and for most brands it is not, because of what follows from it. If brand pages are the primary citation surface, then the competition for a citation is a competition between brand pages. And the inputs most brands use to write those pages are inputs every one of their competitors also has. Four sources sit behind most ecommerce content briefs: the existing product page, Search Console, competitor pages, and general category knowledge. All four are public. Any competitor can reach them. So can any AI writing tool pointed at the category. The result is convergent content: pages that differ in wording and agree in substance. An assistant choosing between five convergent pages has no reason to prefer yours. The material that cannot be copied is not public. It sits in four first-party records that most brands already hold and almost none use for content: support tickets, product reviews, returns reasons, and on-site search queries. Every one of them is readable through a documented API today. This report sets out what AI assistants cite, why public inputs produce interchangeable pages, where the non-public material lives, and a four-question audit that can be applied to a content brief before anything is written. It also states plainly what the evidence does not support, because several claims circulating in this space are wrong. --- ## 1. What AI assistants actually cite Three findings frame the argument. Brand sites are the majority of citations. Profound analysed 11.84 billion citations between 16 April and 16 July 2026, across eight models (ChatGPT, Claude, Google AI Mode, Google AI Overviews, Gemini, Grok, Microsoft Copilot and Perplexity), classifying 3.02 million domains covering 98.3% of global citation volume. More than one citation in two goes to a brand site, with the overall figure at approximately 57%. Social and UGC platforms take a much smaller and highly variable share. Google AI Overviews cites social sources at 15.3% and AI Mode at 14.4%, while Microsoft Copilot uses a social source roughly once in every 29 citations. One caveat matters when reading that figure: "brand site" means any company-operated property, not necessarily the brand being asked about. A citation to a retailer or a manufacturer both count. And Profound has not disclosed how its prompt set was selected. Reddit is the most-cited single domain, and that means less than it sounds. Profound's earlier study with Reddit, published 10 November 2025 across four billion citations, found Reddit to be the single most-cited domain in AI answers at 3.11% share. It ranks first on a very long tail. Rank one is not dominance, and on Copilot the same domain sits around 31st. Ranking no longer determines citation. Ahrefs examined 863,000 keyword SERPs and four million AI Overview URLs, published 2 March 2026. Only 38% of AI Overview citations came from pages ranking in the top ten, down from 76% a year earlier. Put the three together and the picture is clear enough. The citation surface is mostly brand-owned. Getting there is decreasingly a function of where you rank. Which leaves the content itself doing the work, and raises the question of where that content comes from. --- ## 2. The four public inputs, and why each one fails The failure here is structural rather than qualitative. These inputs do not produce bad pages. They produce the same pages. The existing product page. The most common starting point for a rewrite, and the most obviously circular. Your product page is public, indexed, and already read by every competitor in your category. A page rewritten from itself adds phrasing, not substance. Search Console. Records queries that produced an impression or click in Google Search. It cannot record a prompt typed into an assistant, because that string never reaches Google. A brief built from Search Console describes how people searched Google in the past, and the demand it describes is visible to every competitor through any keyword tool. Competitor pages. The most explicitly convergent input available. Content built by surveying what rivals cover produces, by construction, the intersection of what rivals cover. It guarantees parity and forecloses differentiation. General category knowledge. What an experienced writer or a language model already knows about the category. Available to everyone by definition, and the input an AI writing tool defaults to when given nothing else. The compounding problem is that these are also the four inputs most AI content tools use. A brand writing from public inputs with an AI tool is not competing with four rivals doing the same. It is competing with four rivals running similar tools over the same source material, at higher volume. --- ## 3. The four non-public inputs Four records exist inside most ecommerce businesses that no competitor can read. Support tickets. Every question a buyer could not answer from your site, in their own words, already answered once by your team. Gartner's self-service study, published 19 August 2024 with 5,728 respondents, found 43% of customers could not find content relevant to their issue. The tickets are what that failure looks like on your store specifically. Product reviews. Partly public once published, which weakens them, but the corpus in your review platform includes structured attributes, private feedback and unpublished submissions that a competitor scraping your public reviews does not get. Returns reasons. A record of buyers who decided with incomplete information. The reason code names the product and the failure mode together. Native data is thin. Shopify's ReturnReason enum offers a closed list of colour, defective, not as described, size too large, size too small, style, unwanted, wrong item, other and unknown, with returnReasonNote capped at 255 characters and the customer note at 300. Some third-party platforms carry free text: Loop Returns exposes a free-text reason through its warehouse reporting endpoint. On-site search queries. The closest available proxy for a prompt. Buyers typing in their own words, about your actual catalogue, on a property you own. The highest-value subset is the queries that returned nothing. None of these is exotic. What is unusual is using them to decide what a page should say. --- ## 4. The machine-readability proof The objection to all of this is practical: interesting in principle, inaccessible in practice. It does not hold. Every source above is readable through documented interfaces today. | Record | Access | Notes | |---|---|---| | On-site search | Shopify Web Pixels API, search_submitted event | Carries event.data.searchResult.query and the returned productVariants. An empty variants array is a zero-result query. | | On-site search, reporting | Shopify Search & Discovery app | "Searches with no results" report. Trailing 30 days only. | | Returns | Shopify Admin GraphQL, ReturnReason / ReturnLineItem | Closed enum plus returnReasonNote (255 chars) and customer note (300 chars). | | Returns, free text | Loop Returns warehouse reporting endpoint | Free-text return_reason. It is a warehouse report, not a general returns list. | | Support tickets | Gorgias, Zendesk, Front, Re:amaze APIs | All expose conversation content and tagging. | | Reviews | Yotpo, Judge.me, Loox APIs | Loox API access is on paid tiers only. | Two honest notes. WooCommerce has no native returns object, so returns data sits in whichever RMA plugin the store runs and access varies. And we were unable to verify Okendo's API capabilities, because their documentation is client-rendered and could not be read — so no claim is made about it here. The point of this table is not that the integration is trivial. It is that the barrier is attention, not access. --- ## 5. What the evidence says about specific text Two peer-reviewed results point the same way, and both are worth reading precisely rather than as headlines. Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024. Across 10,000 queries, adding citations, quotations and statistics raised source visibility by roughly 27%. Keyword stuffing lowered it by 8.8%. The widely quoted "up to 40%" figure is a best case for particular categories, not the average, and the study tested general web pages rather than ecommerce product pages. Filandrianos et al., "Bias Beware", EMNLP 2025. Across six models and five product sets, social-proof phrasing raised LLM recommendation rates by 9.75% to 22.38%, while scarcity and exclusivity copy cut them by 13% to 46%. Note carefully what was tested: the researchers injected social-proof phrasing, not real customer reviews. It is a proxy, and a counter-intuitive one: the urgency language that works on humans appears to work against you with a model. Both findings converge on the same conclusion. Specific, evidence-bearing, verifiable text performs. Promotional and vague text does not. First-party records are the cheapest available supply of the former, because a returns log and a support queue are made of nothing but specifics. --- ## 6. The provenance audit Four questions, applied to a brief before writing begins. 1. Source. Name the record this page draws on. If the answer is "the category" or "competitor pages", the source is public. 2. Counterfactual. Could a competitor with only public data produce this page? If yes, it will not differentiate you. 3. Specificity. Does the page carry a fact, a figure or a quotation that exists only in your records? 4. Destination. Is this at the right level, product, collection or site-wide? A page at the wrong level fails even with a good source. A page that fails question two is re-briefed, not deleted. Category pages and buying guides need to exist. The audit does not sort content into keep and bin; it sorts content into what might earn a citation and what will merely be present. The audit runs in about a minute per brief, and it runs before the expensive part. Every other quality gate in common use, whether similarity scoring, AI detection or editorial review, runs after the page has been written and paid for, and none examine the source at all. --- ## 7. What this report does not claim Several claims circulate in this area that the evidence does not support. We are not making them. We do not claim AI prefers UGC to brand sites. The largest dataset says the opposite. Any figure showing UGC beating brand-owned domains for citation share should be treated with suspicion. We do not claim Reddit accounts for around 60% of ChatGPT citations. That figure describes share of prompt responses at a peak, immediately before it fell to roughly 10%. We do not claim generic content performs worse than publishing nothing. No study exists showing that, in AI retrieval or in search. Ahrefs' July 2026 analysis of around 331,000 pages found high-AI-signal pages received two to three times fewer impressions and about nine percentage points lower indexation, but that is correlational and rests on a proprietary detector. The defensible claim is that generic content is out-competed and under-indexed, not that it is harmful. We do not claim Google penalises AI-generated content. The helpful content system was folded into core ranking in March 2024. Google's spam policy, updated 15 May 2026, targets scaled content abuse and intent to manipulate rankings, not the use of AI as a tool. We do not claim product reviews measurably change what assistants recommend. No study tests real reviews against commercial assistants. The EMNLP result above is the nearest proxy and it tested injected phrasing. We do not claim a cost saving from support-ticket deflection. The attribution chain does not close, and the figures in circulation do not survive inspection. Stating these limits is not throat-clearing. The argument in this report is structural, and a structural argument does not need inflated numbers to stand up. Content written from a source your competitors cannot reach is difficult for them to reproduce. That holds whether or not anyone has yet measured the citation lift. --- ## Conclusion The question most brands ask about AI discovery is what to write. It is the wrong question, or at least the second one. Brand-owned pages are the majority of what assistants cite. Ranking has come apart from citation. That combination puts the weight on the content itself, and content written from public inputs is, structurally, content every competitor can produce. The first question is what the page is made of. Four records inside your business are invisible to competitors, readable through documented APIs, and currently used for operations rather than content. Reading them as content briefs is not a technology problem. It is a matter of deciding that the source of a page matters as much as its quality. Ask the counterfactual question before you commission anything. It costs a minute, and it is the only gate in common use that runs before the money is spent. For the companion argument on output rather than input, see The Coherence Report. --- ## References 1. Profound, "Where do AI citations come from?", 30 July 2026. 11.84bn citations, 3.02m domains, 16 April to 16 July 2026, 8 models. 2. Profound with Reddit, AI citation analysis, 10 November 2025. 4bn citations. 3. Ahrefs, "AI Overview citations from top-10 pages", 2 March 2026. 863k SERPs, 4m AI Overview URLs. 4. Ahrefs, AI-content analysis, 27 July 2026. ~331k pages, 100k SERPs. Correlational; proprietary detector. 5. Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024. 10,000 queries. 6. Filandrianos et al., "Bias Beware", EMNLP 2025. 6 models, 5 product sets. 7. Gartner, customer self-service survey, 19 August 2024. n=5,728. 8. Google Search, spam policies, page updated 15 May 2026. 9. Shopify, Web Pixels API search_submitted standard event; Search & Discovery analytics; Admin GraphQL ReturnReason. 10. Loop Returns, warehouse reporting endpoint documentation. --- ================================================================================ # The AI Commerce Protocol Guide: agents.md, llms.txt, UCP and the New Plumbing of Discovery URL: https://geoffy.ai/whitepapers/ai-commerce-protocol-guide Last updated: 2026-07-22 ================================================================================ ## Executive summary In under a year, AI commerce has grown its first real protocol layer: discovery files at the site root, structured commerce attributes in the feed, agent-facing endpoints, and crawler-permission tokens. Platforms have moved from ignoring these to shipping them by default. This guide maps the layer as it stands in July 2026 — what each protocol does, what the defaults leave empty, and the boundary that matters most: protocols decide whether an agent can find and transact with you; they say nothing about whether you're the answer it gives. ## 1. Why protocols suddenly matter Consumer behaviour moved first. By April 2026, 77% of US consumers had used AI when shopping in the past six months, and 44% start purchase journeys inside an assistant (Exploding Topics, reported by Search Engine Land). Yet only 2% complete the purchase inside the AI — the assistant is the shop window, not the till. The industry's plumbing has been racing to catch up with exactly that split: standards for being read by agents have matured fastest, standards for transacting through them are still finding their footing — Walmart's public step back from checkout-inside-chat, after conversion ran at roughly a third of its own site's rate, is the defining data point. For a merchant, that resolves the priority question. The protocols worth adopting today are the ones that make you legible at the discovery layer, where the behaviour already is. ## 2. The discovery files Four files at the root of a store now do the work of routing AI crawlers and agents. Each serves a different reader; all must agree. robots.txt — the oldest and still the gatekeeper. It advertises the sitemap and sets crawler permissions (see §4). Every AI fetcher checks it first; a misconfigured robots.txt silently undoes everything below. llms.txt — proposed in late 2024 as a short, model-facing summary of what a site contains: the most important pages, top products, essential context. Its long-form sibling, llms-full.txt, carries enough detail that a model can answer basic questions about the store without crawling further. agents.md — instructions for AI agents rather than models reading passively: what the store is, what's allowed, where to look, how to interact. The significant development of 2026 is that Shopify now ships this family natively: production stores serve llms.txt, llms-full.txt and agents.md at the root by default, with agents.md describing itself as the canonical agent-facing file and llms.txt as a mirror. A /.well-known/ucp endpoint sits alongside them. The default is the opportunity. Pull the default agents.md from any set of Shopify stores and you find the same short generic template — a handful of fixed sections with only the store name and domain substituted in, its largest block routing shopping agents through platform checkout. No product authority. No brand voice. No curated answers about what the brand sells or what an agent should say when a shopper asks. A clothing brand, a supplements brand and a homewares brand ship the same file on day one. The platform has made the file canonical; filling it with substance — an identity statement an agent could quote, canonical product and category URLs, real answers in the brand's voice — remains entirely the merchant's job, and almost no merchant has done it. (Override mechanics are theme-level; check current Shopify documentation for the supported route.) ## 3. The commerce protocols Universal Commerce Protocol (UCP) — announced at NRF in January 2026 and developed with ecosystem partners including Shopify, UCP is Google's framework for plugging brands into AI shopping conversations across Search, Gemini and AI Mode: conversational brand agents on the search surface, offers surfaced at the moment of stated intent, and discovery endpoints a store exposes. Its significance is less any single feature than the admission it encodes: on Google's own surfaces, the shopping journey is expected to happen inside conversations. Merchant Center conversational attributes — the structured-feed side of the same shift. New attribute sets feed Google's conversational layer with the machine-readable signal it needs to answer confidently. Worth adopting with one caution: the feed powers the Shopping graph — carousels and shopping modules — which is not the same retrieval path as conversational citations in chat answers. Feed quality lifts one; it doesn't purchase the other. Checkout rails — instant-checkout integrations inside assistants exist and continue to evolve, but the consumer-trust evidence (31% of US consumers say they wouldn't authorise an AI agent to spend on their behalf) and the Walmart result argue against making them the priority. Discovery on the assistant; transaction on your own surface, where the account, saved payments and returns flow live. ## 4. Crawler permissions: two decisions, not one The most common protocol mistake we see is treating "AI bots" as one category in robots.txt. There are two separate decisions: The training decision. Crawlers like GPTBot, CCBot and Bytespider collect content for model training. Blocking them is a defensible intellectual-property posture and does not remove you from AI answers today. Two tokens are widely misunderstood here: Google-Extended and Applebot-Extended are training opt-out signals, not fetchers that serve answers — Google states explicitly that Google-Extended does not affect a site's inclusion in Search. The retrieval decision. Fetchers such as PerplexityBot and OAI-SearchBot read pages live, at the moment a user asks, so the engine can cite them. Blocking these makes you invisible to the answer surfaces you most likely want traffic from. Audit your robots.txt — and any CDN-level bot management — against both lists separately, and re-check after September 2026, when Cloudflare's announced split of crawler controls into Search, Agent and Training categories lands and existing blanket settings inherit new semantics. ## 5. The job no protocol does Here is the boundary this guide exists to draw. Every protocol above answers a findability question: can an agent locate your store, read it, and (eventually) transact with it. Adopt them all, perfectly, and one question remains entirely open: when a shopper asks for the best product in your category, does the assistant name you? That is decided by what the agent finds when the protocols deliver it to your content — whether your product information is specific, structured, current, and above all consistent with itself across every surface the model reads. A perfect agents.md pointing at contradictory product pages is a well-signposted road to a shop the assistant won't recommend. The evidence base for that claim — including the controlled studies showing structure without substance moves nothing — is assembled in our companion report, The Coherence Report. Protocols make you findable. What makes you chosen is the coherence of what they find. ## 6. Adoption checklist In priority order, for a store starting today: confirm robots.txt and CDN bot settings implement the training/retrieval split deliberately (§4); check what your platform already serves at /llms.txt, /llms-full.txt and /agents.md, and replace default substance with real substance; keep all discovery files consistent with each other and with the visible site; adopt the Merchant Center conversational attributes if you're on Google surfaces; treat checkout-in-chat as an experiment, not a priority; and put your effort where the protocols can't help — the consistency and specificity of the product information itself. ## Conclusion Protocol layers reward early, correct adoption and punish blanket defaults. In 2026 the discovery files are canonical, the commerce protocols are converging, and the defaults are empty. The merchants who fill them — and whose underlying catalogue deserves the citation the plumbing makes possible — are competing for answer slots most of their category doesn't yet know exist. --- Geoffy builds coherence infrastructure for AI commerce: discovery files, structured data and intent-tier content engineered on the merchant's own domain and kept in sync as the catalogue changes. Live today on Shopify and WordPress / WooCommerce. ## References - Exploding Topics survey (n=1,009 US consumers), reported by Search Engine Land, April 2026. - Walmart EVP product remarks on ChatGPT checkout performance, reported March 2026. - Shopify production-store discovery files: first-party probes, May–July 2026. - Google, Universal Commerce Protocol announcement, NRF, January 2026. - Google Search documentation on Google-Extended. - Cloudflare, crawler-control category changes effective 15 September 2026. ================================================================================ # The Coherence Report: Why AI Recommends Brands That Agree With Themselves URL: https://geoffy.ai/whitepapers/the-coherence-report Last updated: 2026-07-22 ================================================================================ ## Executive summary Through 2024 and 2025, advice on AI visibility ran ahead of evidence. That gap closed materially in May 2026, when two controlled studies — one measuring what doesn't move AI retrieval, one measuring what captures citations — were published within ten days of each other. Read together, they frame the discipline: structure alone has no measurable effect, and coherent brand-owned pages win the citation. This report assembles that evidence, the concentration data that raises the stakes, and the mechanism that explains both. ## 1. The two findings that frame 2026 The null result. In May 2026, Ahrefs published a controlled study of schema-only optimisation: 1,885 treated pages against 4,000 matched controls, evaluated with a difference-in-differences design. The measured effect on AI retrieval — +2.4% in Google's AI Mode, +2.2% in ChatGPT — was, in the authors' words, statistically indistinguishable from zero. Adding structured data markup to otherwise unchanged pages did not move whether AI systems retrieved or cited them. The authors were careful to carve out what schema still does: it feeds entity recognition and knowledge graphs — the upstream registration layer. What it does not do is substitute for the page itself being worth citing. The capture result. The same month, BrightEdge measured where consideration-stage citations actually go. Across eight industries, brand-owned commercial pages captured between 42% and 79% of citations at the consideration stage of the buying journey. The popular assumption — that AI answers are built almost entirely from third-party editorial and forum content — does not survive contact with this data. When a brand's own page carries consistent, structured, checkable information, engines cite it directly. One study shows the shortcut doesn't work. The other shows the prize for doing the real work. The variable separating them is not markup, budget, or domain authority. It is whether the brand's information is coherent enough for a model to stake an answer on. ## 2. The mechanism: why contradiction reads as risk An AI assistant answering "what's the best magnesium for sleep?" has to compose a confident recommendation in a few hundred tokens, usually with no prior context. To do that it reads what it can find — the product page, structured data, an FAQ, a marketplace listing, a review page — and resolves everything into one description. If those sources agree, the model can describe the product specifically and commit. If they disagree — one dosage on the product page, another in an old blog post; one positioning on the homepage, another in the feed — the model faces a choice it is built to avoid: assert something that might be wrong. So it hedges, or it omits, and recommends the competitor next door whose sources don't disagree with themselves. This is why the Ahrefs result should surprise nobody. Schema is a promise about the page. If the page, the feed, and the wider web don't keep that promise consistently, the markup adds a claim without adding confidence. And it is why the BrightEdge result is the encouraging half: the model prefers an owned page it can trust, because a coherent primary source is the lowest-risk citation available. Earlier academic work pointed the same way. The Princeton-led GEO study (ACM KDD 2024) found structured, explicitly attributed content increased visibility in generative responses by up to 40% against unstructured equivalents — structure with substance, measured together, not markup alone. ## 3. The concentration effect: no long tail in an answer Coherence would matter less if AI answers were generous. They are not. In benchmark testing of roughly 3,800 supplement-category prompts run against five engines between January and April 2026 (2026 Supplements AI Visibility Index, 5WPR), four brands captured an estimated 47% of all observed citations. In separate, narrower category testing published earlier (Avenue Z, September 2025, three engines), a substantial minority of tested brands appeared zero times — not ranked low; absent. Search never behaved this way. The results page had ten links, then more below, then a second page; being twelfth was survivable. An answer has three to five names in it. The competition is winner-take-most, and it resolves faster than search's slow rank decay — because there is nothing below the fold of a sentence. Concentration also compounds the coherence effect in both directions. Brands the models can describe confidently get cited, which produces more consistent third-party coverage, which strengthens the next retrieval. Brands the models skip generate no such record. The gap widens without anyone deciding it should. ## 4. Fragmentation makes coherence the only portable strategy A year ago, a plausible response was: learn one engine's quirks and optimise for it. That assumption has expired. ChatGPT's share of generative AI traffic fell from roughly 76% to roughly 53% in a year (Similarweb), while Gemini's share more than doubled and Perplexity and Copilot grew into real volume. Discovery is not consolidating onto one engine; it is splintering across several, each retrieving differently and each changing frequently. Per-engine tactics therefore don't scale — four engines, each moving every few weeks, cannot each be gamed indefinitely. What transfers across all of them is the property none of them can ignore: information that agrees with itself wherever it's read. Coherence is engine-independent. It pays off in every engine at once, including those that haven't launched yet. ## 5. What coherent brands do differently Across first-party visibility audits — including a sustained audit programme in the supplements category — the brands that perform in AI answers share observable habits rather than budgets: One story, everywhere. The product page, the FAQ, the reviews vocabulary and the structured data use the same claims and the same language — often because a single voice wrote them and kept them matched. Comparison-ready specificity. Attributes stated explicitly (dose, form, material, certification), in terms a model can check against a competitor, rather than dissolved into lifestyle copy. Maintained, not launched. Prices, availability and claims that stay synchronised as the catalogue changes. Coherence decays by default; the winners treat it as infrastructure with upkeep, not a project with an end date. A recurring, uncomfortable pattern: brands with the deepest content libraries — years of SEO investment, agency retainers, topic clusters — are routinely beaten in AI answers by smaller competitors whose entire site is a handful of crisp, repeating, machine-readable claims. Search rewarded depth. AI rewards agreement. They are different optimisation targets, and success at the first confers nothing at the second. ## 6. Implications For brands, the order of work follows from the evidence. First, look: ask three engines the question your best customer asks and read the answers side by side — presence, accuracy, consistency. Second, reconcile: find where your own surfaces contradict each other, because that is what the model sees. Third, structure what's true: schema and discovery files built from a coherent catalogue, not painted over an incoherent one. Fourth, maintain: schedule the checking, because drift is the default state of any live catalogue. For the industry, the May 2026 studies should retire two comfortable narratives at once: that AI visibility is a markup checkbox, and that it is entirely at the mercy of third-party coverage. The evidence says the decisive surface is the one the brand already owns — held to a standard most catalogues were never built to meet. ## Conclusion The judge has changed. Search's judge tolerated internal contradiction because it ranked pages against each other; the assistant's judge cannot, because it has to say something true in two sentences. Structure gets you read. Coherence gets you chosen. The brands that internalise the difference earliest will spend the next two years being the answer — and the rest will spend them wondering where the demand went. --- Geoffy builds coherence infrastructure for AI commerce: structured discovery on the merchant's own domain, kept in sync as the catalogue changes. Live today on Shopify and WordPress / WooCommerce. ## References - Ahrefs, Schema markup and AI retrieval: a controlled study, 11 May 2026. - BrightEdge, consideration-stage citation analysis across eight industries, 21 May 2026. - 5WPR, 2026 Supplements AI Visibility Index (Jan–Apr 2026 testing). - Avenue Z, supplement AI visibility testing, September 2025. - Similarweb, generative AI traffic share data, 2025–2026. - Aggarwal et al., GEO: Generative Engine Optimization, ACM KDD 2024. ================================================================================ # GEO for Agencies: Building AI Discovery as a Service URL: https://geoffy.ai/whitepapers/geo-for-agencies Last updated: 2026-07-22 ================================================================================ A strategic guide for agencies ready to build a GEO practice. > Updated July 2026. The client conversation has moved from “is this real?” to “how do we do it?”: 77% of US shoppers have used AI to shop in the past six months, and 44% start purchase journeys inside an assistant (Exploding Topics/Search Engine Land, April 2026). Meanwhile usage is fragmenting — ChatGPT's share of generative AI traffic has fallen from roughly three-quarters to about half in a year (Similarweb) — which makes single-engine audits, still the norm among agencies, an incomplete service. Cross-engine coverage is now the credibility bar. ## Executive Summary AI product discovery is emerging as the next major ecommerce visibility channel. For agencies, it represents a significant service opportunity: clients need guidance, the expertise gap is wide, and the ongoing nature of AI discovery management makes it a natural retainer. This paper outlines why GEO belongs in the agency offer, what a credible GEO service looks like, and how to build a practice that compounds over time. ## Section 1: Why this moment matters for agencies The ecommerce agency landscape has been shaped by waves of channel emergence — organic search, paid search, paid social, influencer, CRO. Each wave created a window of opportunity for agencies that moved early to establish expertise, build case studies, and capture long-term client relationships before the channel became commoditised. AI product discovery is in that early window now. AI-referred retail traffic grew dramatically through 2024 and 2025. The channel is still small as a percentage of total ecommerce traffic, but it's growing faster than any channel at the equivalent stage in its development. And the traffic quality is exceptional — higher conversion rates, longer sessions, and higher average order values than most channels agencies currently manage. For agencies, the question is not whether to build a GEO practice, but how quickly. ## Section 2: What clients need and why they can't get it themselves Ecommerce clients are aware of AI. Most have asked assistants product questions themselves. Many have noticed that their products don't appear in AI recommendations. What they lack is a clear path forward. The challenge is genuine. Building effective AI discovery infrastructure requires understanding how AI retrieval systems work, how to structure product data for machine reasoning, how to make the merchant's existing product and category pages answer at every depth of intent, and how to keep all of it current as the catalogue changes. It's not a simple content exercise. Most in-house ecommerce teams don't have the bandwidth, the expertise, or the tooling to build this well. That's the agency opportunity. ## Section 3: What a GEO service looks like A credible agency GEO service has four components. Discovery audit. Understanding the client's current AI visibility position — which products are showing up in AI recommendations, which aren't, and where the structural gaps are. This is a valuable standalone deliverable that creates urgency and defines the scope of work. Discovery infrastructure build. Creating the first-party discovery layer on the client's domain — structured pages organised around the real questions their customers ask AI assistants, covering the catalogue at the right level of depth and breadth. Ongoing management. Keeping discovery infrastructure current as the catalogue evolves. This is where the retainer lives. Products change, new lines are added, stock status shifts — the page content and its structured data need to stay accurate and aligned. Reporting and strategy. Tracking AI referral trends, coverage growth, and product-page performance. Translating directional data into strategic recommendations that demonstrate value and guide further investment. ## Section 4: How to position GEO to clients The framing that works is not technical. It's competitive. Your products are either showing up in AI recommendations or they're not. Your competitors are either building AI discovery infrastructure or they're not. The channel is growing. The early advantage is real. What are we doing about it? That's a conversation every ecommerce client is ready to have — and most don't have an agency partner currently equipped to answer it. Avoid leading with methodology or technology. Lead with the business outcome: recommendation visibility in a channel where the traffic quality is exceptional and the infrastructure advantage compounds over time. ## Section 5: The tooling question Agencies building a GEO practice face a build-versus-buy decision on tooling. Building proprietary infrastructure is possible but expensive, slow, and difficult to maintain across multiple client catalogues. The ongoing nature of drift management and catalogue sync alone represents a significant ongoing engineering overhead. Geoffy is designed for agency use. Multi-client catalogue management, first-party publication on each client's own domain, and structured reporting are built into the platform. Agencies use Geoffy as the infrastructure layer and focus their own value-add on strategy, audit, and client management. ## Conclusion The agencies that build GEO practices now will have a structural advantage in the ecommerce market for years. The window of differentiation is open. The client need is real. The channel is compounding. The question for agency leaders is not whether GEO belongs in the offer. It's whether to build the capability now, while the advantage is meaningful, or later, when it's expected. Talk to us about the Geoffy agency programme → ## About Geoffy Geoffy is a Generative Engine Optimisation platform that helps ecommerce brands and the agencies that serve them build first-party AI discovery infrastructure — structured, maintained, and built to compound over time. ## Related reading - How Agencies Can Offer GEO as a Service — the practitioner-facing version of this argument. - How to Choose an AI-Visibility Tool — a neutral buyer's guide to the tooling category. - Optimising for ChatGPT Is Already the Wrong Goal — why single-engine auditing under-serves clients. - Solutions for agencies — how Geoffy supports an agency practice. ## References Adobe Analytics / Digital Commerce 360 — AI referral growth and quality data (2025–2026). Aggarwal et al. — GEO: Generative Engine Optimization (ACM KDD 2024). Cloudflare Radar 2025 — AI crawling activity trends. All data reflects publicly available research as of March 2026 and should be treated as directional. ================================================================================ # The Future of Product Discovery URL: https://geoffy.ai/whitepapers/future-of-product-discovery Last updated: 2026-07-22 ================================================================================ > Updated July 2026. Two developments since publication sharpen this paper's central claim. First, the discovery and transaction layers have visibly separated: Walmart publicly stepped back from checkout-inside-chat after conversion ran at roughly a third of its own site's rate, refocusing on being present in the conversation while completing purchases in its own environment. Second, consumer data confirms the same split — most shoppers now use AI to decide, and almost none to buy in-channel (Exploding Topics/Search Engine Land, April 2026). The decision engine is here; the till has stayed put. ## Introduction Every decade or so, the way people find products changes. The history of ecommerce discovery is, in many ways, a history of interfaces. Each era has been defined by the system sitting between consumer intent and the products they buy. Those systems have changed dramatically, but one pattern stays consistent: brands that adapt early build durable advantages. We are now in the early stages of another transition. AI assistants are becoming a meaningful interface between consumers and commerce. The scale of change AI-referred retail traffic grew 35x in the 11 months from July 2024 to May 2025, then doubled again between February and May 2025 alone. [1] This paper examines how discovery evolved through three eras, what AI-mediated discovery means in practice, and what the medium-term future of ecommerce visibility looks like. ## The three eras of ecommerce discovery Discovery has moved through three distinct technological eras, each defined by a different mechanism for connecting consumer intent to products. Era Period Primary interface Core model Web directories Pre-2000 Human-curated category lists Manual browsing to find products Search engines 2000–2023 Keyword-driven search Ranked links, user compares options AI assistants 2024–present Conversational questions Synthesised answers and shortlists As of early 2026, Google AI Overviews appear in an estimated 25% of searches. When present, organic click-through rates can fall sharply year-on-year. [2] ## From search engine to decision engine Search engines help users locate options. AI assistants increasingly act as decision engines. In a search journey, users compare many links themselves. In an AI journey, comparison happens inside the model and users are shown a short list — a handful of products. Dimension Search engine AI assistant User input Short keyword query Natural language question with context System output Ranked links Synthesised answer with recommendations Comparison work Done by the user Done by the AI system first Products surfaced Many options Small shortlist Optimisation target Ranking position Structured interpretability and attribute completeness The strategic implication is straightforward: the competitive surface is now narrower, and missing the shortlist can mean being excluded from consideration entirely. ## How AI systems evaluate products Large language models do not rank pages the way traditional search engines do. They reason over product information. When users ask product questions, AI systems evaluate: - attribute match - use case fit - comparative positioning - trust and consistency - recency and availability Research finding A Princeton-led GEO study found structured, explicitly attributed content increased visibility in generative responses by up to 40% versus unstructured alternatives. [4] This is why parity matters. Structured data that contradicts visible content is a trust failure. ## Looking ahead: AI-native commerce The current phase is likely transitional. Several trends point toward AI-native commerce: - conversational shopping as a first-class journey - multi-turn recommendation sessions - agentic purchasing behaviours - stronger platform-level integrations and feed pathways OpenAI Operator and Anthropic Computer Use are early demonstrations of agentic workflows in which a machine navigates and evaluates the buying journey on a user’s behalf. [5] For merchants, this raises the bar for machine-readable product quality. ## Strategic implications for ecommerce brands AI discovery does not replace SEO, but it changes what visibility means. Strategic question Search-era answer AI discovery-era answer How do I get found? Rank for keywords on page one Ensure complete, crawlable, structured product meaning How do I cover intent? Target keyword variants Answer Broad/Mid/Ultra intent on the product page How do I measure progress? Rankings, impressions, CTR AI referral trends plus directional visibility testing Long-term moat Links and domain authority Attribute depth, intent breadth, semantic consistency AI referral traffic is still small in absolute share, but quality signals are strong: higher conversion rates, longer sessions, and higher revenue per visit in cited retail datasets. [1] ## Conclusion The future of discovery is conversational. The core question is no longer only “Can users find my pages?” but “Can AI systems understand and recommend my products?” Generative Engine Optimisation is the infrastructure response to that shift. ## About Geoffy Geoffy is a Generative Engine Optimisation platform that helps ecommerce brands on Shopify and WordPress make product catalogues legible to AI assistants and AI search systems. Geoffy generates intent-tiered content and structured JSON-LD, published onto the merchant’s own canonical product pages. ## About the author Anthony Gale is Co-Founder of Geoffy and has spent more than two decades working in ecommerce, digital growth, and product development. ## Related reading - AI Is a Discovery Channel, Not a Checkout Channel — the consumer data behind the split. - AI Found the Product. Walmart Took the Cart Back. — what happened when a retailer tested checkout-in-chat at scale. - What is Answer Engine Optimisation (AEO)? — the adjacent vocabulary. - What is Generative Engine Optimisation (GEO)? — the foundational explainer. ## References 1. Adobe Analytics / Digital Commerce 360 — AI referral growth and quality metrics (2025–2026). 2. Conductor / Pew Research (2025) — AI Overviews prevalence and CTR impact. 3. Profound — Typical LLM citation breadth (domains cited per response). 4. Aggarwal et al. (Princeton / Georgia Tech / IIT Delhi) — GEO: Generative Engine Optimization (ACM KDD 2024). 5. Industry coverage of OpenAI Operator and Anthropic Computer Use (2025). All data reflects publicly available research as of March 2026 and should be treated as directional. ================================================================================ # The Shift from Search to AI Discovery URL: https://geoffy.ai/whitepapers/shift-from-search-to-ai-discovery Last updated: 2026-07-22 ================================================================================ > Updated July 2026. The behavioural data has continued in this paper's direction: 77% of US consumers have used AI when shopping in the past six months, with 44% starting purchase journeys inside an assistant (Exploding Topics/Search Engine Land, April 2026). One nuance has strengthened since March — the shift is not to a single AI engine but to several at once, with ChatGPT's share of generative traffic falling from roughly three-quarters to about half in a year as Gemini, Perplexity and Copilot grow. Optimising for “AI” now means optimising for coherence across engines, not for one engine's quirks. ## Executive summary The discovery model is changing. For more than twenty years, ecommerce discovery has been dominated by search engines. Brands optimised product pages, category pages, and content to rank in search results. When customers searched, search engines returned a list of links. The customer clicked, compared, and eventually chose what to buy. That model is now changing, and the data is unambiguous. Key Stat AI-referred retail traffic grew 35x in the 11 months from July 2024 to May 2025, doubling every two to three months. It is now the fastest-growing traffic channel in ecommerce by a significant margin. [1] Instead of searching, customers are increasingly asking AI assistants for direct answers: - "What's the best running shoe for flat feet?" - "What's a good espresso machine under £500?" - "Which standing desk is best for small spaces?" These systems do not return ten links. They return recommendations. And those recommendations are shaped by how well product information is structured, not just how well pages are ranked. This shift introduces a new optimisation discipline: Generative Engine Optimisation (GEO). For ecommerce brands, the window to build early infrastructure is now. ## Section 1: The end of the ten blue links Traditional SEO was built around one core idea: ranking. Search engines crawled the web, indexed pages, and ranked them based on signals such as keywords, backlinks, technical structure, and site authority. The outcome was a list of search results. The user performed the work of comparing options across multiple sites. AI assistants fundamentally change this experience. Instead of returning links, they produce synthesised answers. A user asking for "the best lightweight hiking backpack" may receive a direct recommendation with a small set of products already evaluated and summarised. In this model, the discovery process happens inside the AI system. The websites that benefit are the ones whose products are understood well enough to be recommended, not just indexed. Context As of February 2026, Google's AI Overviews appear in an estimated 25% of searches. When an AI Overview is present, organic click-through rates fall 61% year-on-year, and only 1% of users click a link within the overview. The traffic model has structurally changed. [2] The scale of consumer adoption is accelerating faster than most brands anticipated. AI shopping behaviour Finding Source AI assistants replacing search 36% of generative AI users now replace traditional search with AI for product research. Adobe Analytics, 2025 [1] Primary product research tool 72% of active AI platform users rely on AI as their primary tool for researching products and brands. Adobe Analytics, 2025 [1] Millennial adoption 46% of Millennials now use AI-assisted shopping; over 50% for higher-income groups. Adobe Analytics, 2025 [1] Revenue per visit growth Revenue per visit from AI referrals grew 254% year-on-year during the 2025 holiday season. Digital Commerce 360, 2026 [3] Conversion quality Holiday 2025 AI referral conversions were 31% higher than non-AI traffic, double the advantage seen in 2024. Digital Commerce 360, 2026 [3] The implication for ecommerce brands is straightforward: the channel is small today, but it is compounding rapidly and the traffic that does arrive is high-quality. ## Section 2: Why AI systems choose some products and ignore others Large language models do not rank pages in the same way as search engines. Instead, they try to understand information. When evaluating product content, AI systems look for signals that help them interpret product attributes, relationships between products, categories and variants, and suitability for specific use cases. In other words, AI systems build semantic models of products and intent. This is why traditional SEO tactics alone are not enough. A product page that ranks well in Google might still be opaque to an AI assistant trying to synthesise a recommendation. Research finding A Princeton-led study on GEO found that structured content combining statistical evidence and citation-backed claims increased AI visibility by up to 40% versus unstructured equivalents. Structure is not a nice-to-have; it is a primary signal. [4] There is also a category dimension: - Consumer electronics and home goods currently lead in AI-referred traffic share. [1] - High-consideration, high-price products show the strongest correlation between AI referrals and direct conversion. [1] - "Best X for Y" and "X under £Z" queries are among the highest-volume AI shopping prompts. Adobe Analytics data from the 2025 holiday season also showed AI-referred visitors spent 45% more time on site, viewed 13% more pages per visit, and had a 27% lower bounce rate than non-AI traffic. [3] ## Section 3: The four layers of AI product discovery In building Geoffy, we have found that effective AI discovery requires four distinct layers. Each is necessary. None is sufficient alone. ### Layer 1: Discovery signalling AI systems first need to know where content exists. Signals such as sitemaps, structured product feeds, llms.txt files, and internal linking help crawlers locate and prioritise content. A Cloudflare analysis found that user-action AI crawling increased more than 15x in 2025. [5] Signalling alone does not guarantee visibility, but its absence guarantees invisibility. ### Layer 2: Structured product understanding Once content is discovered, AI systems must interpret it. This requires clear signals about product attributes, variants and relationships, categories and use cases. Structured formats such as JSON-LD and semantic product graphs give AI systems the data they need to reason about fit for a specific question. ### Layer 3: Intent-aligned content on the pages you already own When AI systems answer shopping questions, they rarely reference a single product page. They summarise clusters of products related to a specific intent. That means the merchant has to be able to answer questions such as: - "Best standing desks under £500" - "Lightweight travel backpacks for carry-on" The answer belongs on the product and category pages the merchant already publishes — readable for humans, structured for machines, and on the URLs that already carry the site's authority. Building a parallel set of AI-facing pages splits the catalogue into two versions that drift apart, and moves the content away from the pages that earned the trust. ### Layer 4: Attribution and measurement One of the biggest challenges of AI discovery is attribution. AI-referred traffic often appears as direct or unattributed in standard analytics. Without structured landing experiences and intent-specific measurement, merchants cannot demonstrate GEO ROI or optimise over time. ## Section 4: Why ecommerce infrastructure needs to evolve Most ecommerce platforms were designed for traditional search. A typical stack includes product feeds, SEO plugins, content tools, and standard analytics. Useful, but not designed for AI-mediated discovery. Common gaps include: - product descriptions written for human skimming, not machine reasoning - inconsistent or incomplete attribute fields - little or no mapped intent coverage - no AI-first discovery sitemap layer Industry assessment Jeremy Moser (uSERP) has noted that roughly 80% of effective GEO practice is strong, structured content fundamentals. What is new is applying this at intent-cluster level, not just individual product-page level. [6] There is valid scepticism in the market around inflated GEO claims. Geoffy's approach is infrastructure-first: - no guaranteed visibility claims - no hidden AI-only content - no cloaking - parity between visible page content and machine-readable outputs ## Section 5: How Geoffy approaches AI discovery infrastructure Geoffy is built on a single principle: product data should be explicit, verified, and maintained — not generated and hoped for. Most GEO implementations fail not because the concept is wrong, but because the execution is fragile. Attributes get invented rather than grounded. Structured data drifts out of sync with visible content. Page content goes stale as catalogues change. Each of these failures degrades trust signals and reduces recommendation inclusion over time. Geoffy's platform handles the full lifecycle: building discovery infrastructure from your catalogue, keeping it current as products change, and maintaining parity between what AI systems read and what customers see. For ecommerce teams, the outcome is a first-party AI discovery layer on their own domain — structured, maintained, and built to compound over time. See how Geoffy works → ## Section 6: The SEO analogy and where it breaks down Comparing GEO today to early SEO is directionally useful. Early movers in search earned compounding advantages. The same pattern is appearing in AI discovery. Growth comparison AI-referred sessions grew 527% in the first half of 2025. By comparison, one of the fastest mobile ecommerce growth periods (2013–2014) was around 40% year-on-year. AI discovery is moving an order of magnitude faster. [7] Where the analogy breaks down is measurement. AI influence often happens in sessions without direct referral clicks, so GEO today is best understood as infrastructure investment, not direct-response media. ## Section 7: Is your catalogue AI-ready? Use this practical assessment to identify gaps. Assessment area Question to ask Why it matters Attribute completeness Do products have structured, consistent attributes beyond title and description? AI systems need explicit attributes to reason about fit. Intent coverage Do you have pages that answer "best X for Y" questions in your category? Discovery happens at intent level, not just product-page level. Structured data Do pages use JSON-LD that matches visible content? Schema remains a key hybrid retrieval signal. AI crawlability Do you provide llms.txt and an AI-facing sitemap signal? Crawlers need clear discovery paths. Drift management Do pages stay accurate as stock, prices, and specs change? Stale outputs degrade trust and recommendation quality. Attribution tracking Can you identify AI-assistant referrals and intent paths? Without measurement you cannot optimise or prove ROI. If the answer to most of these is no or not sure, there is a meaningful gap. The fix is not a platform rebuild. It is structured implementation and maintenance. ## Conclusion The move from search results to AI answers is still early, but the direction is clear. AI assistants, recommendation engines, and conversational search are already shaping product discovery behaviour. As agentic commerce matures, brands whose products are structured, verified, and legible to machines will have a compounding advantage. Geoffy was built on a simple principle: product data should be explicit, deterministic, and trustworthy. The infrastructure brands build in the next 12–24 months will shape their AI discovery position for years. ## About Geoffy Geoffy is a Generative Engine Optimisation platform that helps ecommerce brands on Shopify and WordPress make product catalogues legible to AI assistants and AI search systems. Geoffy generates intent-tiered content and structured JSON-LD, published onto the merchant's own canonical product pages, with no hidden content, no cloaking, and full parity between visible copy and machine-readable data. - geoffy.ai - linkedin.com/company/geoffy-geo - crunchbase.com/organization/geoffy ## About the author Anthony Gale is a co-founder of Geoffy. He has spent more than two decades working in ecommerce, digital growth, and product development, from early-stage startups to global retailers. ## Related reading - SEO vs GEO: How AI Product Discovery Is Changing Ecommerce Visibility — the two disciplines side by side. - Optimising for ChatGPT Is Already the Wrong Goal — why the shift is plural, not singular. - The Citation Layer Is Getting Smaller — what narrowing citation behaviour means for brands. ## References 1. Adobe Analytics — The explosive rise of generative AI referral traffic / Q2 2025 AI referral insights. 2. Conductor (Nov 2025) and Pew Research Center (Jul 2025) — AI Overviews appearance rates and CTR impact. 3. Digital Commerce 360 — Generative AI online holiday shopping traffic 2025 (Jan 2026). 4. Aggarwal et al. (Princeton / Georgia Tech / IIT Delhi) — GEO: Generative Engine Optimization, ACM KDD 2024. 5. Cloudflare Radar Year in Review 2025 — AI user-action crawling growth. 6. Jeremy Moser, uSERP — Digiday: GEO hype analysis (Mar 2026). 7. Previsible 2025 AI Traffic Report — AI-referred session growth. All data cited reflects publicly available research as of March 2026. AI discovery remains a rapidly evolving field, and metrics should be treated as directional. ================================================================================ # Why Product Data Integrity Is the Foundation of AI Discovery URL: https://geoffy.ai/whitepapers/product-data-integrity-for-ai-discovery Last updated: 2026-07-22 ================================================================================ Structure gets you into the game. Integrity keeps you there. > Updated July 2026. Two findings published since this paper strengthen its argument from different directions. Ahrefs' controlled study (May 2026: 1,885 pages against 4,000 matched controls) found schema-only optimisation produced retrieval effects statistically indistinguishable from zero — structure without integrity moves nothing, exactly as argued below. And BrightEdge (May 2026) measured brand-owned commercial pages capturing 42–79% of consideration-stage citations across eight industries — the payoff side: when the owned page is coherent, it is the citation. Integrity remains the differentiator between the two results. ## Abstract Most GEO implementations are more fragile than they appear. Getting structured, answer-ready content and matching schema live on your canonical product pages is the first step — but without rigorous data integrity, the same implementation that improves visibility in the short term can degrade it over time. This paper examines why data integrity is the critical requirement for sustained AI recommendation inclusion. ## Section 1: The problem with "generate and publish" The most common GEO failure pattern we observe is this: a team structures its product pages and the schema behind them, publishes, and sees initial improvement. Then, over weeks and months, the catalogue changes — prices shift, products go out of stock, specifications are updated, variants are discontinued. The pages don't keep up. What was accurate becomes stale. What was consistent becomes contradictory. AI systems are sensitive to this. When the information they find about a product is internally inconsistent — when what's stated in one place doesn't match what's visible in another — trust signals degrade. Recommendation inclusion follows. Data integrity isn't a nice-to-have quality measure. It's the mechanism by which AI visibility is maintained over time. ## Section 2: What integrity means in practice At its core, data integrity for AI discovery means one thing: what AI systems read about a product must match what customers actually see. This sounds simple. In practice, it requires deliberate systems. Product data changes constantly — and in most ecommerce catalogues, it changes faster than any manual content process can keep pace with. Prices change daily. Stock status shifts by the hour. Specifications are updated when suppliers change formulations. Variants are added and removed. Each of these changes creates a potential gap between what's declared and what's true. And in AI recommendation systems, gaps create risk. ## Section 3: The integrity requirements that matter most From building and maintaining AI discovery infrastructure across ecommerce catalogues, three integrity requirements stand out as most consequential. Attribute grounding. Every attribute that appears in a discovery context must be traceable to a verifiable source. Invented or inferred attributes — however plausible they seem — create a category of trust failure that compounds over time. If a system claims a product has a certain specification that doesn't appear in the source catalogue data, and an AI assistant later encounters contradictory information, the damage isn't just to that product — it extends to the domain's overall credibility as a source. Parity between structured and visible content. AI systems increasingly cross-reference machine-readable signals against the human-readable content on the same page. Mismatches — structured data claiming something the visible page doesn't confirm — are a reliability signal. They suggest either a technical error or deliberate manipulation, neither of which supports confident recommendation. Drift management. Catalogues are live data systems. The content and structured data layered onto them must be treated the same way. Pages that were accurate at publication can become misleading within days as the underlying catalogue shifts. Without systematic monitoring and update processes, that layer has a natural decay curve. Integrity degrades silently, and visibility follows. ## Section 4: Why this changes the evaluation of GEO tooling When evaluating GEO approaches — whether that's a platform, an agency implementation, or internal development — data integrity should be a primary evaluation criterion, not a secondary one. The questions worth asking are straightforward: How does this approach ensure that attributes are grounded rather than generated? What happens when a product changes — how is discovery content kept current? Is there a mechanism to detect and resolve discrepancies between structured outputs and visible page content? The answers to these questions predict long-term performance far better than the quality of initial implementation. ## Section 5: The infrastructure-first principle At Geoffy, we describe our approach as infrastructure-first rather than content-first. The distinction matters. A content-first approach to GEO treats this as a content production exercise: write the copy, publish it, move on. An infrastructure-first approach treats the same content and its structured-data mirror as live data outputs — connected to source truth, validated before publication, and maintained as the underlying data changes. The difference isn't immediately visible in the outputs. A product page produced either way might look identical on day one. The difference becomes apparent over time, and particularly under the pressure of a live, changing catalogue. Brands that build on infrastructure principles compound their AI visibility advantage. Brands that treat GEO as a content project tend to see initial gains followed by gradual erosion. ## Conclusion AI discovery is a live system, not a campaign. The ecommerce brands that will build durable recommendation visibility are the ones that treat their discovery infrastructure with the same rigour they apply to their product data: verified, maintained, and built to stay accurate as the underlying truth changes. Structure is the entry requirement. Integrity is what keeps you in the game. ## About Geoffy Geoffy is a GEO platform built on infrastructure-first principles. The platform is designed to keep discovery outputs verified, current, and aligned with source product truth — across catalogues of any size, at any pace of change. ## About the author Branko Goricnik is CTO and Co-Founder of Geoffy. He has spent his career building data infrastructure and platform architecture for ecommerce and digital products. ## Related reading - SEO Rewards Depth. AI Rewards Coherence. — the same argument from the audit floor. - Four Supplement Brands Own Half the AI Shortlist — what concentration looks like in one category. - How AI Assistants Choose Products to Recommend — the mechanics behind the selection. - The Anatomy of a GEO-Optimised Store — the structural checklist. ## References Aggarwal et al. — GEO: Generative Engine Optimization (ACM KDD 2024). Cloudflare Radar 2025 — AI crawler growth and retrieval behaviour. Adobe Analytics (2025) — AI referral traffic quality trends. All data reflects publicly available research as of March 2026. ================================================================================ # The GEO Playbook for Ecommerce URL: https://geoffy.ai/whitepapers/geo-playbook-for-ecommerce Last updated: 2026-07-22 ================================================================================ > Updated July 2026. Since this playbook was published, the discovery-file layer described in Capability 4 has gone platform-native: Shopify now ships agents.md and llms.txt at the root of every store by default — generic templates a brand must still fill. Consumer research now puts AI firmly at the discovery end of the journey (77% of US shoppers have used AI to shop in the past six months, yet only 2% check out inside it — Exploding Topics/Search Engine Land, April 2026), and usage is fragmenting across engines rather than consolidating on one. The framework below stands; the urgency has moved. ## Executive summary Product discovery is shifting from search results to AI-generated recommendations. Consumers are increasingly asking assistants direct product questions. Those systems shortlist options based on structured product understanding, not page ranking alone. Why this matters now AI-referred retail traffic grew 35x between July 2024 and May 2025, with stronger engagement and conversion quality than many traditional channels. [1] This playbook outlines five capabilities commerce teams need, the common mistakes to avoid, and a practical implementation checklist. ## The AI discovery funnel Traditional path: Keyword → Search results → Click → Product page → Purchase AI discovery path: Question → AI synthesis → Recommendation → Click → Purchase In AI discovery, the assistant performs the comparison step before the user clicks. ## The five capabilities ecommerce stores need # Capability What it means in practice 1 Structured product data Consistent machine-readable attributes across your catalogue. 2 Intent-driven content depth The pages you already publish, answering real “best X for Y” and “X under £Z” queries. 3 AI-readable content Clarity-first copy that supports machine reasoning. 4 Discovery infrastructure XML sitemap, llms.txt, JSON-LD, internal linking, crawlability. 5 Attribution and measurement AI referral channel visibility and intent-level tracking. ## Capability 1: Structured product data AI systems reason over explicit attributes. Vague marketing language is weak signal. Attribute Unstructured (weak) Structured (AI-ready) Material Made with premium materials primary_material: recycled nylon Use case Perfect for travel and commuting usage_environment: travel, commuting Dimensions Compact and lightweight dimensions: 44×30×15 cm, weight: 0.8kg Compatibility Works with most laptops max_laptop_size: 15-inch Always ground attributes in source data. If unknown, mark as not provided. ## Capability 2: Intent-driven content depth Single product pages answer “what is this product?”. AI users often ask “what should I buy for this situation?”. Use a three-tier intent model: - Broad: category exploration - Mid: contextual fit - Ultra: precise purchase constraints Cover all three on the pages you already publish — the product page carrying the depth for its own SKU, the category page doing the group-level version. They are blocks of content, not destinations of their own. Resist the obvious shortcut of building a separate page per buying question. It splits the catalogue into two descriptions that drift apart the moment stock or pricing moves, and it shifts the content off the URL that already holds your links, reviews and purchase history — asking an engine to cite your weakest page instead of your strongest. ## Capability 3: AI-readable content AI-readable content is explicit and specific. - include concrete attributes - include use-case fit and constraints - avoid vague promotional language as primary signal ## Capability 4: Discovery infrastructure Core infrastructure requirements: - XML sitemap coverage for every product and category page - llms.txt at domain root - JSON-LD parity with visible content - internal links between related product and category pages - server-side render for AI-facing content Crawling trend Cloudflare Radar reported AI “user-action” crawling growth above 15x in 2025. Discovery infrastructure is now table stakes. [4] ## Capability 5: Attribution and measurement AI influence is broader than direct referral clicks. Track what is measurable now: - custom GA4 channel groups for ChatGPT, Gemini, Claude, Perplexity, Copilot - intent-specific landing pages - crawler log patterns Treat current metrics as directional while the channel matures. ## Common mistakes to avoid Mistake Why it fails What to do instead Treating GEO as standard SEO AI systems are not ranking links the same way. Focus on intent coverage and attribute completeness. Relying only on ad feeds Assistants often retrieve from web content and relationships. Publish first-party content on your own product and category pages. Skipping structured data parity Mismatches reduce trust signals. Enforce schema-to-visible-content parity. Publishing hidden AI-only content Creates trust and policy risk. Keep transparent, visible-first outputs only. The brands that start now will be better positioned as AI discovery matures. If you'd like to explore what GEO infrastructure looks like for your catalogue, Geoffy is a good place to start. ## Conclusion The channel is still early, but the direction is clear. Brands that build AI-readable product infrastructure now will have a structural advantage as recommendation systems mature. ## About Geoffy Geoffy helps ecommerce teams turn product catalogues into AI-readable discovery infrastructure, including intent-tiered pages and structured outputs with visible-content parity. ## About the author Anthony Gale is Co-Founder of Geoffy and has worked across ecommerce, digital growth, and product-led infrastructure for more than two decades. ## Related reading - The Anatomy of a GEO-Optimised Store — the practical companion to the five capabilities. - The Ninety-Second AI Visibility Test — the “where do I start” answer. - Shopify Picked agents.md. The Default Is Bland. — a Capability 4 update. - What is Answer Engine Optimisation (AEO)? — the adjacent vocabulary. ## References 1. Adobe Analytics / Digital Commerce 360 — AI referral growth and quality data. 2. Profound — LLM citation breadth benchmarks. 3. Aggarwal et al. — GEO: Generative Engine Optimization (ACM KDD 2024). 4. Cloudflare Radar 2025 — AI crawling activity trends. 5. Conductor (2025) — AI referral share benchmarks. 6. Enrich Labs (2026) — GEO adoption trend observations. All data reflects public sources available as of March 2026 and should be interpreted as directional. ================================================================================ # Geoffy documentation URL: https://geoffy.ai/docs Last updated: 2026-09-16 ================================================================================ Integration guides for making your product catalogue readable by AI search engines — on a custom-built storefront, on Shopify, or on WordPress. ================================================================================ # FAQ URL: https://geoffy.ai/docs/faq Last updated: 2026-09-16 ================================================================================ Questions that span every platform — what a crawler actually sees, what Geoffy does and does not publish, and how verification and publishing behave. ================================================================================ # Custom-built storefront URL: https://geoffy.ai/docs/headless Last updated: 2026-09-16 ================================================================================ Integrate Geoffy into your own Next.js or Astro storefront — with Shopify behind it, or with no commerce platform at all. ================================================================================ # Astro quickstart URL: https://geoffy.ai/docs/headless/astro Last updated: 2026-09-16 ================================================================================ Add Geoffy to an Astro storefront — the widget, structured data in the head, the root files, the namespace and your robots.txt, in seven steps. ================================================================================ # Checking it worked URL: https://geoffy.ai/docs/headless/astro/concepts/checking-it-worked Last updated: 2026-09-16 ================================================================================ Geoffy fetches your live page and looks for the markup. What each pending reason means, and how to run the same checks yourself. ================================================================================ # How updates arrive URL: https://geoffy.ai/docs/headless/astro/concepts/how-updates-arrive Last updated: 2026-09-16 ================================================================================ Server-rendered Astro refreshes within the revalidate window. A fully static build has no accelerated path at all — read this before choosing. ================================================================================ # Prove you own your domain URL: https://geoffy.ai/docs/headless/astro/concepts/prove-you-own-your-domain Last updated: 2026-09-16 ================================================================================ Serve one verification file, run the check, and every Geoffy artifact endpoint starts answering. Until then they all return 404. ================================================================================ # Where to put the markup URL: https://geoffy.ai/docs/headless/astro/concepts/where-to-put-the-markup Last updated: 2026-09-16 ================================================================================ Server-rendering decides whether a crawler reads you. Astro also lets you put the structured data in the head, which is a mild bonus rather than a requirement. ================================================================================ # Which kind of site is yours URL: https://geoffy.ai/docs/headless/astro/concepts/which-kind-of-site-is-yours Last updated: 2026-09-16 ================================================================================ Where your catalogue comes from is the only thing that differs between the two kinds of custom-built site — and it decides how you set up. ================================================================================ # The IndexNow key URL: https://geoffy.ai/docs/headless/astro/guides/indexnow-key Last updated: 2026-09-16 ================================================================================ The namespace only authorises what is under it. Your product pages are not, so they need the key at your site root — and skipping this fails silently. ================================================================================ # Testing locally URL: https://geoffy.ai/docs/headless/astro/guides/local-testing Last updated: 2026-09-16 ================================================================================ See the widget, the structured data and your merged robots.txt before any of it is on your live site — and know which two things a local run cannot tell you. ================================================================================ # The Geoffy namespace URL: https://geoffy.ai/docs/headless/astro/guides/namespace Last updated: 2026-09-16 ================================================================================ One rest-parameter endpoint puts every other Geoffy surface on your own domain. It needs on-demand rendering, and prerender = false is not optional. ================================================================================ # The product page URL: https://geoffy.ai/docs/headless/astro/guides/product-page Last updated: 2026-09-16 ================================================================================ Call getGeoffyProductMarkup, send the structured data to the head, and place the widget where a shopper should see it. ================================================================================ # Crawler rules URL: https://geoffy.ai/docs/headless/astro/guides/robots-txt Last updated: 2026-09-16 ================================================================================ Append the Geoffy block to your own robots.txt. Where it goes in the file is the part that actually costs you something. ================================================================================ # The site-wide files URL: https://geoffy.ai/docs/headless/astro/guides/root-files Last updated: 2026-09-16 ================================================================================ llms.txt, llms-full.txt and agents.md, as three endpoints under src/pages — all three written out, not "repeat as needed". ================================================================================ # Troubleshooting URL: https://geoffy.ai/docs/headless/astro/guides/troubleshooting Last updated: 2026-09-16 ================================================================================ Every known failure in this integration, indexed by the symptom you actually see rather than by the cause. ================================================================================ # No commerce platform URL: https://geoffy.ai/docs/headless/astro/recipes/no-commerce-platform Last updated: 2026-09-16 ================================================================================ A worked example — an Astro content-collection catalogue with no Shopify behind it, where Geoffy builds its catalogue by crawling your own pages. ================================================================================ # Shopify behind your storefront URL: https://geoffy.ai/docs/headless/astro/recipes/shopify-behind-your-storefront Last updated: 2026-09-16 ================================================================================ A worked example — Astro in front, Shopify Storefront API behind, Geoffy reading the same handle both use. ================================================================================ # Artifact endpoints URL: https://geoffy.ai/docs/headless/astro/reference/artifact-endpoints Last updated: 2026-09-16 ================================================================================ The public API behind every helper — what each endpoint returns, where it lands on your own domain, and how it behaves when Geoffy is degraded. ================================================================================ # Dashboard settings URL: https://geoffy.ai/docs/headless/astro/reference/dashboard-settings Last updated: 2026-09-16 ================================================================================ Every setting Geoffy stores for a custom-built site, what it feeds, and the two fields that no longer exist. ================================================================================ # The package URL: https://geoffy.ai/docs/headless/astro/reference/package Last updated: 2026-09-16 ================================================================================ Every export of @geoffy/headless/astro and the shared client, with signatures and the options they all take. ================================================================================ # Next.js quickstart URL: https://geoffy.ai/docs/headless/nextjs Last updated: 2026-09-16 ================================================================================ Add Geoffy to a Next.js App Router storefront — the widget, the structured data, the root files, the namespace and your robots.txt, in seven steps. ================================================================================ # Checking it worked URL: https://geoffy.ai/docs/headless/nextjs/concepts/checking-it-worked Last updated: 2026-09-16 ================================================================================ Geoffy fetches your live page and looks for the markup. What each pending reason means, and how to run the same checks yourself. ================================================================================ # How updates arrive URL: https://geoffy.ai/docs/headless/nextjs/concepts/how-updates-arrive Last updated: 2026-09-16 ================================================================================ Your pages refresh on their own within the revalidate window. Mounting one route makes it seconds instead — and skipping it is a real option. ================================================================================ # Prove you own your domain URL: https://geoffy.ai/docs/headless/nextjs/concepts/prove-you-own-your-domain Last updated: 2026-09-16 ================================================================================ Serve one verification file, run the check, and every Geoffy artifact endpoint starts answering. Until then they all return 404. ================================================================================ # Where to put the markup URL: https://geoffy.ai/docs/headless/nextjs/concepts/where-to-put-the-markup Last updated: 2026-09-16 ================================================================================ Server-rendering decides whether a crawler reads you. Position on the page does not — and in Next.js it is not a choice you have anyway. ================================================================================ # Which kind of site is yours URL: https://geoffy.ai/docs/headless/nextjs/concepts/which-kind-of-site-is-yours Last updated: 2026-09-16 ================================================================================ Where your catalogue comes from is the only thing that differs between the two kinds of custom-built site — and it decides how you set up. ================================================================================ # The IndexNow key URL: https://geoffy.ai/docs/headless/nextjs/guides/indexnow-key Last updated: 2026-09-16 ================================================================================ The namespace only authorises what is under it. Your product pages are not, so they need the key at your site root — and skipping this fails silently. ================================================================================ # Testing locally URL: https://geoffy.ai/docs/headless/nextjs/guides/local-testing Last updated: 2026-09-16 ================================================================================ You can see the widget, the structured data and your merged robots.txt before any of it is on your live site — and three things will otherwise waste your afternoon. ================================================================================ # The Geoffy namespace URL: https://geoffy.ai/docs/headless/nextjs/guides/namespace Last updated: 2026-09-16 ================================================================================ One catch-all route puts every other Geoffy surface on your own domain — which is where the citations they earn should land. ================================================================================ # The product page URL: https://geoffy.ai/docs/headless/nextjs/guides/product-page Last updated: 2026-09-16 ================================================================================ Mount GeoffyProduct, pass the right handle, and pass canonicalUrl so the component knows which page it is on. ================================================================================ # Crawler rules URL: https://geoffy.ai/docs/headless/nextjs/guides/robots-txt Last updated: 2026-09-16 ================================================================================ Append the Geoffy block to your own robots.txt. Where it goes in the file is the part that actually costs you something. ================================================================================ # The site-wide files URL: https://geoffy.ai/docs/headless/nextjs/guides/root-files Last updated: 2026-09-16 ================================================================================ llms.txt, llms-full.txt and agents.md, as three route handlers — and why a next.config.js rewrite loses a race you cannot see. ================================================================================ # Troubleshooting URL: https://geoffy.ai/docs/headless/nextjs/guides/troubleshooting Last updated: 2026-09-16 ================================================================================ Every known failure in this integration, indexed by the symptom you actually see rather than by the cause. ================================================================================ # No commerce platform URL: https://geoffy.ai/docs/headless/nextjs/recipes/no-commerce-platform Last updated: 2026-09-16 ================================================================================ A worked example — a Next.js catalogue with no Shopify behind it, where Geoffy builds its catalogue by crawling your own product pages. ================================================================================ # Shopify behind your storefront URL: https://geoffy.ai/docs/headless/nextjs/recipes/shopify-behind-your-storefront Last updated: 2026-09-16 ================================================================================ A worked example — Next.js App Router in front, Shopify Storefront API behind, Geoffy reading the same handle both use. ================================================================================ # Artifact endpoints URL: https://geoffy.ai/docs/headless/nextjs/reference/artifact-endpoints Last updated: 2026-09-16 ================================================================================ The public API behind every helper — what each endpoint returns, where it lands on your own domain, and how it behaves when Geoffy is degraded. ================================================================================ # Dashboard settings URL: https://geoffy.ai/docs/headless/nextjs/reference/dashboard-settings Last updated: 2026-09-16 ================================================================================ Every setting Geoffy stores for a custom-built site, what it feeds, and the two fields that no longer exist. ================================================================================ # The package URL: https://geoffy.ai/docs/headless/nextjs/reference/package Last updated: 2026-09-16 ================================================================================ Every export of @geoffy/headless/next and the shared client, with signatures and the options they all take.