How to rank in ChatGPT

You cannot rank in ChatGPT. There is no position two. When a shopper asks for the best trail shoe under £120, ChatGPT names three or four products and the rest do not exist. So the question is not how to rank. It is how to get named.

ChatGPT names a product when it can read the product, trust the facts and match them to the question. Which gives you seven things to fix on a product page. Here they are, in the order that matters.

How ChatGPT chooses a product

ChatGPT with search on does three things in about four seconds. It rewrites the shopper’s question into two or three searches. It reads the top results, plus any pages it already trusts from training. Then it picks the products whose facts fit the question and writes the answer, citing the pages it used.

So a product gets named when three conditions hold.

1. The page is in the results ChatGPT reads, which is mostly Bing’s index plus known sources.
2. The page states the facts the question needs, in a form the model can lift without guessing.
3. Those facts agree with every other place the model has seen the product.

Ranking in Google helps with the first. The second and third are where product pages fail.

The seven checks

1. State what the product is in one plain sentence. “The Arch Support Runner 2 is a road running shoe for flat feet and overpronation.” Not “Engineered for your journey.” The model needs a sentence it can quote.
2. State who it is for and what it beats. ChatGPT answers “best for” questions. Give it the answer. “Best for flat feet. Firmer arch than the Pegasus, lighter than the Kayano.”
3. Put the price, stock and sizes in text, not just in a widget. If the price only exists in JavaScript, the model may never see it. Say it in plain text and repeat it in the schema.
4. Make the schema agree with the page. Product JSON-LD with name, price, currency, availability, brand and the key attributes. If the schema says £109 and the page says £119, the model hedges or skips you.
5. Keep it consistent everywhere. Your feed, your resellers, your marketplace listings. The model has seen all of them. Three materials for one product means no material.
6. Give the engines a front door. An llms.txt at your domain root that lists your products and what they are. A grounding file per product if you can. ChatGPT does not require them. It uses them when they exist.
7. Only ground what you can sell. Sold out products should drop out of the machine readable layer. Naming a product the shopper cannot buy is worse than not being named.

Before and after

Same product, same facts, two shapes.

Before

Meet the Arch Support Runner 2. Engineered for runners who demand more, it combines responsive cushioning with a supportive feel that keeps you going mile after mile. Available now.

After

Arch Support Runner 2 is a road running shoe for flat feet and overpronation.
Support: medial post, 10mm heel drop
Weight: 285g (UK 9)
Price: £109 inc. VAT, free UK delivery
Sizes: UK 6 to 13, in stock
Best for: runners with flat feet doing 20 to 50km a week
Compare: firmer arch than the Nike Pegasus, lighter than the ASICS Kayano

The first version says nothing a model can use. The second answers “best running shoe for flat feet under £120” without the model having to infer a thing. Both can sit on the same page. The first for the shopper, the second for the engine.¹

How to measure it

Turn on the AI Assistant channel in GA4. It groups sessions from ChatGPT, Perplexity, Gemini and Claude. Note the conversion rate and average order value against organic search. On the stores we have seen data for, AI sessions convert at two to three times organic and carry a bigger basket.²

Then ask ChatGPT the same ten buying questions once a month and record who gets named. That is your share of answers. It moves within weeks of fixing the pages.

Doing this at scale

The seven checks take about twenty minutes per product by hand. At 50 products that is a week. At 2,000 it is a job nobody finishes, and the answers drift again after the next model update.

Geoffy runs the seven checks on every in stock product in a Shopify or WordPress store, publishes the machine readable layer, asks the engines, and corrects what drifts. Available now

Questions people ask

Does ChatGPT use Google rankings?
Not directly. ChatGPT search leans on Bing’s index and on sources it already trusts. A page that ranks well in Google usually ranks well in Bing, so good SEO helps, but it is not the deciding factor.

Can I pay to appear in ChatGPT?
Not for organic answers. OpenAI has tested ads and shopping integrations. The product answers shoppers get today are chosen by the model from what it can read.

Does ChatGPT read schema markup?
Yes, when it fetches the page. Product JSON-LD is the cleanest way to give it price, availability and attributes. It has to match the page text.

How long does it take?
Facts show up in answers within days of the page being re-read. Share of answers builds over one to three months. Measure from day zero.

What about Perplexity and Google AI Overviews?
The same seven checks work for all of them. Perplexity cites more sources per answer, so it is often the first to name you.