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.
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.
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.
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 discovery pages
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 discovery increasingly happens through intent-aligned collections such as:
- “Best standing desks under £500”
- “Lightweight travel backpacks for carry-on”
These pages should exist on the merchant domain, be readable for humans, and structured for machines.
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
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. Discovery pages go 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.
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.
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 GEO bundles, structured JSON-LD outputs, and first-party discovery pages published directly on the merchant’s domain, 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.
References
- Adobe Analytics — The explosive rise of generative AI referral traffic / Q2 2025 AI referral insights.
- Conductor (Nov 2025) and Pew Research Center (Jul 2025) — AI Overviews appearance rates and CTR impact.
- Digital Commerce 360 — Generative AI online holiday shopping traffic 2025 (Jan 2026).
- Aggarwal et al. (Princeton / Georgia Tech / IIT Delhi) — GEO: Generative Engine Optimization, ACM KDD 2024.
- Cloudflare Radar Year in Review 2025 — AI user-action crawling growth.
- Jeremy Moser, uSERP — Digiday: GEO hype analysis (Mar 2026).
- 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.