Why AI names other brands instead of yours

For twenty years, shopping online started the same way. A shopper typed a few words into Google, got a page of links and opened two or three of them. Your product page was one of the pages they might open, and once it was open, it did the talking.

Now it starts with a question. The shopper types it into an AI model, such as ChatGPT, Gemini, Perplexity, Claude or Google’s AI Overviews, the five this guide follows. One answer comes back, naming a few brands with a reason for each, and the shopper can buy from that list without opening anything else. A store that isn’t on it never gets the chance its product page used to have.

An AI model names the products that the pages it reads give it a reason to name. A store can be left out in two ways, because its own pages do not hand that reason over or because the other sites the AI model reads describe someone else. Two questions about one product tell you which.

The examples follow Tarrowby, a made-up store that sells travel bags. The other brands and websites are real, and what they are shown saying is invented. The product is the Overnighter 30, a backpack for work trips of one to three nights.

What happens between the question and the answer?

The AI model searches before it writes. It runs a few searches of its own, opens some of the pages they turn up and builds its answer from what those pages say. It also draws on what it learned in training, but for a question about what to buy it usually searches as well, since products and prices keep changing.

Take a shopper who asks ChatGPT, What’s a good backpack for a three-day work trip? OpenAI says (opens in a new tab) ChatGPT turns a question like this into narrower searches of its own, and Google says (opens in a new tab) its AI features do the same. Here the searches might be for business travel backpacks, for reviews of work-trip backpacks and for backpacks that open like a suitcase.

One question, followed to its answerExample

1A shopper asksWhat’s a good backpack for a three-day work trip?

2The AI model runs searches of its ownbest backpack for business travel 2026work trip backpack reviewsbackpack that opens like a suitcase

3It opens a handful of the pages it findsOutsideYouTubeAmazonQuorapeakdesign.comtarrowby.comNever opened

4It writes one answerThree bags come up for short work trips: one from Peak Design, one from Osprey and one from Patagonia.Peak DesignOspreyPatagonia

The searches turn up many pages, and the AI model opens a handful. In the example it reads a review on Outside, a YouTube video, an Amazon listing, a Quora answer and Peak Design’s own site. Then it writes one answer naming Peak Design, Osprey and Patagonia, with a reason for each taken from what it read. It never opened tarrowby.com, so the Overnighter 30 could not be named, though it was made for exactly this trip.

Two things decided that answer. One was which pages got opened, and the other was what those pages said. The next two sections are about getting your pages opened, and the one after is about how much of a page an AI model can read once it does. How AI search decides which brands to recommend explains in full how the searches are chosen and which pages get believed.

My Google rankings are fine. Why doesn’t that carry over?

For Google’s own AI features, good rankings count for a great deal. The other AI models run searches of their own and lean heavily on what other sites say, so a store can rank well on Google and still be missing from their answers.

Google says (opens in a new tab) its AI features are part of Google Search, so making a site do well in them is still SEO. Its guide also mentions answer engine optimisation, or AEO, one of the names people now use for this kind of work. It means getting your products named in the answers AI models give, and it is what this guide is about.

ChatGPT, Claude and Perplexity search in their own ways. ChatGPT sends its searches to partner search companies, and OpenAI names (opens in a new tab) Microsoft and Shopify among them. Claude and Perplexity send their own software to read the pages they search, which the next section covers. A page’s place in Google’s results does not travel with it into theirs.

They also lean on sites the brand does not own. When the question is which product is best, most of the pages an AI model cites are review sites, magazines, retailers, videos and forums. So a store can be first on Google for its own name and still be missing from the answer. Ranking for your name helps a shopper who already knows you, and the shopper in the example had never heard of Tarrowby.

If you sell on Shopify, your products may already be available to ChatGPT and other AI models. Shopify switches this on by default (opens in a new tab) for eligible stores, and chapter 4 looks at what it sends. Being available is the start. It puts a product among the ones an answer can name, and whether it is named still depends on the pages the AI model reads.

Can AI models get into your store at all?

Each AI company sends software round the web to read pages, and a small file on your store tells that software whether it may come in. Checking the file takes a minute.

The software is called a crawler. Each company has more than one, for different jobs, and three of them read pages for search answers. They are OAI-SearchBot for ChatGPT, Claude-SearchBot for Claude and PerplexityBot for Perplexity.

The file is called robots.txt, and every Shopify store has one (opens in a new tab). Type your store’s address followed by /robots.txt into a browser, as in tarrowby.com/robots.txt, and the file opens. Look for the three names. The lines under a name are the rules for that crawler, and Disallow: / under one of them shuts it out of the whole store. If none of the names appears, the general rules apply to them as to any crawler. Lines that keep crawlers away from the cart and checkout are normal.

Each company says what shutting out its search crawler costs. OpenAI says (opens in a new tab) a site that opts out of OAI-SearchBot will not be shown in ChatGPT’s search answers. Blocking Claude-SearchBot, Anthropic says (opens in a new tab), may make a site show up less in Claude’s search results. Perplexity asks (opens in a new tab) site owners to allow PerplexityBot so that their sites can appear in its search results.

If one of them is shut out and nobody meant it to be, the fix is a change to a template in your theme. Shopify warns that a wrong line in this file can cost a store all its traffic, so the change is a job for whoever looks after your theme.

Google asks for nothing extra. Its rule (opens in a new tab) for AI Overviews and AI Mode is that a page must be in Google’s index, its record of the pages it has read, and able to show a snippet. A snippet is the short preview of text under an ordinary search result. If your product page turns up in an ordinary Google search with a line of text under its title, it meets the rule. Three ways an AI model learns about your products lists every crawler and covers the ones that collect pages for training.

How much of your product page can an AI model read?

An AI model reads the text your store sends, and that can be less than the page you see. Every product page exists in two versions. One is the text your store sends when the page is asked for, called the page source. The other is the page your browser builds once your theme and its apps have run their scripts, the small programs that fill in tabs and load reviews.

In December 2024, Vercel and MERJ (opens in a new tab) watched the crawlers of OpenAI, Anthropic and Perplexity at work and found that none of them ran a page’s scripts. Google’s crawler, which Gemini relies on, did. Crawlers change, and that study was one measurement at one time. Whatever they do now, text that arrives with the page reaches every reader, and text that arrives later may not.

Pictures work the same way. Words printed on a product photo, or a comparison chart saved as an image, give an AI model nothing to read. Microsoft tells site owners (opens in a new tab) to keep important answers out of tabs and expandable menus, and never to leave a fact that matters only in an image.

On a Shopify store, the facts at risk usually sit in reviews loaded by an app, in tabs of details filled in after the page loads, and in charts saved as pictures. A coffee grinder made for small kitchens might have its best line, that it is quiet enough to use while the rest of the house sleeps, printed on a photo and nowhere else.

Which of the two gaps is yours?

A store left out of an answer has one of two gaps, and two questions about one product show which. The gaps sit between three things worth keeping apart. There is what is true about your product, which lives in your documents, such as the design spec and the supplier’s pages. There is what your pages say in text an AI model can read. And there is what the answer says.

A gap on your own pages means something true about your product is missing from the text of your pages, or sits only in a picture or a tab that loads late. A gap on other sites means your pages say it plainly and the answer still leaves you out, because the other sites the AI model read describe someone else. The two are fixed in different places, so find out which one you have before changing anything.

Ask for a fact first, and name the product. Naming it sends the AI model looking for that product, so it is likely to open your own pages. If it gets the fact wrong or hedges, your pages are not handing the fact over. Run the two checks above, the robots.txt file and the page source, then turn to chapters 4 and 5, which deal with a gap on your own pages.

If it gets the fact right, ask the shopper’s question. Being named means neither gap is open for that question today. Being left out means the AI model knows your product and still recommends others, so it is leaning on other sites for that answer. That is a gap on other sites, and chapter 6 deals with it.

The test works for any store. A shop selling a rain jacket for cycling to work would first ask whether its jacket’s hood fits over a helmet, naming the jacket, and then ask for a good rain jacket for cycling to work.

Ask each question several times before believing the result. The same question asked twice can bring back different brands, and chapter 2 explains why and how to count across many askings.

What comes next, and in what order?

Five more steps remain. The rest of the guide takes them in the order the work goes in, and each chapter hands the next one something you can use.

1. In chapter 2 you find the questions your shoppers really ask, and count how often AI answers name you and which pages they cite.
2. In chapter 3 you decide what each product should be known for, and which document proves each reason.
3. In chapter 4 you read what your store says about each product, everywhere an AI model may read it, and sort what you find into fix first, add and keep.
4. In chapter 5 you rewrite each fact as a plain sentence that makes sense on its own, in every field that should carry it.
5. In chapter 6 you get other sites to say it, ask them to correct what they get wrong, and keep it all true as your products change.

Two questions told you which gap one product has. Chapter 2 widens that to the questions your shoppers really ask, across every product that matters.