For ChatGPT to recommend your products, it has to be able to read your site, find the data the question asks for and see other sources mention you. In practice: let its crawler in, put the attributes people ask about (weight, drop, material, surface) in the HTML of your product pages, publish buying guides in your categories, because guides are what it cites, and get named on the review sites it relies on. Then measure what it answers, question by question.
How does ChatGPT decide which store to recommend?
When someone asks "where should I buy women's trail running shoes?", ChatGPT searches the web, reads a few pages and answers with a few stores. There's no second page of results: either you're in that answer, or for that shopper you don't exist.
The concentration is strong. In our October 2026 study, with buying questions about sports gear asked to AI assistants with web search, 6 in 10 answers named the same store (58–68%). A mid-size generalist got a 15% share, and 0% on product questions.
What decides who gets in isn't only the brand. From what we see when we measure, the assistant picks the pages that best answer that specific question, and then the stores those pages mention or sell.
Why does ChatGPT recommend your competitors instead of you?
When a store loses a question, it's usually for one of these four reasons:
| Cause | How you spot it | What to do |
|---|---|---|
| A competitor's guide | The answer cites a "how to choose" or "best of" article from another store. | Publish your own guide for that question, linked to what you stock. |
| A review or comparison site | The answer relies on a magazine, a blog or a comparison site that names other stores. | Get those sites to mention you, with correct information about you. |
| Missing product data | The question asks for a detail (weight, drop, surface) your product pages don't have and your competitors' do. | Fill that attribute across the whole category, not on three products. |
| A page AI can't read | Your site never shows up, not even for questions you should win. | Check robots.txt, and that the content is in the HTML, not only in JavaScript. |
The same store can lose each question for a different reason. That's why one fix for everything doesn't work: you have to look at the cause question by question.
Can ChatGPT read your store?
It's the first thing to check, because if this fails, nothing else matters.
- Let its crawler in. OpenAI uses
OAI-SearchBotto surface websites in ChatGPT search, and according to its documentation, sites that block it aren't shown in its search answers. Look at your/robots.txtand check you don't block it, by name or with a generalDisallow. - Put the content in the HTML. According to a December 2024 analysis by Vercel, OpenAI's and Anthropic's crawlers download JavaScript files but don't execute them. If your product page loads the price, description or attributes with JavaScript, they probably don't see them.
- Mark up your products with structured data. Schema.org Product markup with name, brand,
GTIN and an offer (
Offer) with price and availability tells any machine what you sell without it having to interpret your page. Make sure it includes the GTIN: it's what shows your product is the same one other sites mention.
A robots.txt that lets ChatGPT's crawler in can be as simple as this:
User-agent: OAI-SearchBot
Allow: /If you block other AI crawlers so they don't train on your content, do it by name and let the search ones through. They're different things.
What product data does AI need to recommend you?
Whatever the question asks for. People don't ask ChatGPT for "shoes", they ask for shoes for something: for women, for muddy mountain trails, for wide feet. If that detail isn't on your product page, the assistant can't know your product fits, even if it does.
| The question | What it needs to know |
|---|---|
| "Women's trail running shoes with good grip in mud" | Gender, terrain, lug depth, outsole compound, drop, weight |
| "Padel rackets for beginners" | Shape (round, teardrop, diamond), balance, weight, core hardness, recommended level |
| "Football boots for artificial grass" | Soleplate type (AG, TF, FG), surface it's designed for, upper material |
| "Day hiking backpack that fits as cabin luggage" | Capacity in liters, outer dimensions, weight, waterproofing |
Examples of the kind of questions we measure.
Two practical rules. First, the detail has to be written down, not only in a photo or a brand PDF. Second, it has to be on every product in the category. If only three of your forty padel rackets state their balance, to the assistant your store doesn't know padel. We explain it with a before and after example in how to optimize your product pages for AI.
Why do buying guides count for more than product pages?
Because, on product questions, they're what AI cites. In the same study, 54 of 54 store citations on product questions went to buying guides, not product pages. When someone asks "which one should I buy", the assistant looks for a page that compares and explains, and then names the stores that appear in it.
A guide that works for this:
- Answers one specific question, with a title almost identical to that question.
- Gives the answer up front, then explains how to choose, with the criteria that matter in that category.
- Compares options in a table, with data rather than adjectives.
- Links to the products you stock, with product pages that carry the same data.
- Shows a visible update date, and actually gets updated when the range changes.
Guides and product pages reinforce each other: the guide gets you cited, and complete product pages make what the guide promises verifiable.
What role do review and comparison sites play?
On many questions, the assistant doesn't cite any store directly: it cites a magazine, a specialist blog or a comparison site, and names the stores that source mentions. If your competitors appear on those sites and you don't, you lose the question even if your own site is perfect.
What helps here is knowing which sources the assistant cites in your categories, because they aren't always the ones you'd expect. With that list, the work is the usual PR and affiliate work: be on those pages, with correct information about you.
How do you know whether ChatGPT recommends your store?
By asking it, but properly. A single query doesn't tell you much, because the same question can get different answers from one time to the next. The reliable way is to repeat each question several times, from your market and with web search, and look at the share of answers you appear in.
| Checking by hand | Measuring systematically | |
|---|---|---|
| Questions | Whatever comes to mind | The real buying questions in your categories |
| Repeats | One, maybe two | Several per question, to separate change from noise |
| Assistants | The one you use | ChatGPT, Claude and Gemini separately |
| Competitors | The ones you remember | Your share against each store you compete with |
| Cause | Gut feeling | Which page it cited and why it wasn't you |
To start, doing it by hand with ten questions already tells you a lot. To make decisions, you need the systematic method.
Where do I start this week?
- Open your
/robots.txtand check thatOAI-SearchBotcan get in. - Open a product page, view the source and look for the price and the attributes. If they're not there, AI doesn't see them either.
- Write down the ten most important buying questions in your main category and ask ChatGPT each one three times.
- Note which stores it names and which pages it cites. If they're buying guides, you know what you're missing.
- Pick the attribute that comes up most in those questions and check how many of your products have it.
This is what Semantico does systematically: it asks ChatGPT, Claude and Gemini about your categories, tells you which store they recommend and why it isn't you, and fills the missing product data across the category with its catalog builder.
Common questions
Does regular SEO help you show up in ChatGPT?
It helps, but it isn't enough. ChatGPT searches the web and reads some of the pages it finds, so a page that doesn't rank has little chance. But out of what it reads it names only a few stores, and it picks the ones that best answer that specific question: with the data it asks for, in a guide or on a review site that mentions them.
What is GEO?
Generative engine optimization: getting picked up by AI search and shopping assistants such as ChatGPT, Claude or Gemini, the way SEO is about getting picked up by Google.
Does it work the same way with Claude and Gemini?
The principles are the same: being able to read your site, finding the data the question asks for and seeing other sources mention you. But each assistant searches in its own way and can recommend different stores for the same question, so it's worth measuring them separately.
How long does a change take to show?
It depends on when the page is crawled and indexed again, and there's no fixed number. That's why you measure before and after the change, with several samples per question, and compare with questions you haven't touched.
Does ChatGPT give everyone the same answer?
Not exactly. The same question can get different answers from one time to the next, and the app can differ from what the API returns. A single query doesn't tell you much: repeat each question several times and look at the pattern.
Do I need a big store to get recommended?
No. In our October 2026 study in sports retail, 54 of 54 store citations on product questions went to buying guides, not product pages. A specialist store can have the best guides and the most complete product pages in its category.
