How AI assistants decide which stores to mention

Ask ChatGPT, Perplexity or Gemini for a product recommendation and you’ll usually get a short list of stores, sometimes with links. Merchants reasonably want to know how that list gets made. Nobody outside these companies knows every detail, and it changes often. But the broad mechanics are well understood, and most of them are things a store can influence.

Two places answers come from

An AI assistant draws on two very different sources when it answers a shopping question:

• What the model learned in training. Large, well-known brands show up here because they were written about widely. A small store rarely does, and that knowledge is frozen at a cutoff date.
• What it retrieves right now. For product questions, assistants increasingly search the web live, read a handful of pages, and summarize them. This is where an independent store can realistically appear.

The second path is the one you can work on. It behaves a lot like search, because it usually is search: the assistant runs queries against a web index, picks promising results, fetches them and writes an answer from what it read.

What they look for on your pages

Once a page is fetched, the model has to understand it quickly. Pages that are easy to summarize accurately tend to be the ones that get summarized. In practice that means:

• Plain facts in the text. Material, sizes, price, who it’s for, where it ships. If the key details only live in images or tabs loaded by scripts, a crawler may never see them.
• Structured data. Product schema states price, availability, brand and ratings as data rather than prose, which removes guesswork.
• Clear titles and descriptions. The same meta titles and descriptions that help in Google help an assistant decide whether a page is relevant.
• A summary of the whole store. A file like llms.txt gives a model the overview without crawling every page. Here’s what goes in one.

Can they reach you at all?

AI companies run their own crawlers, with user agents such as GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot. Some are used for training and some for live search, and companies document them separately. If your robots.txt blocks the search-related ones, your pages can’t be fetched for answers, however good they are.

That’s a business decision, and blocking training crawlers while allowing search crawlers is a reasonable middle ground for many stores. The mistake is making the decision by accident. Our Shopify robots.txt guide covers how to check.

Signals from beyond your site

When an assistant weighs several candidate stores, it reads what others say too: reviews, comparison articles, directories, forums. A store that’s consistently described the same way across the web is easier to recommend with confidence than one with conflicting or thin information.

You can’t control all of that, but you can:

• Keep your name, policies and key facts identical everywhere they appear.
• Get listed in the directories and data sources relevant to your niche.
• Make reviews easy to leave, and mark them up so they’re machine-readable.

A short checklist

1. Make sure AI search crawlers aren’t blocked in robots.txt (unless you mean them to be).
2. Put the important product facts in page text, not only in images.
3. Turn on Product, Organization and Breadcrumb structured data.
4. Write specific meta titles and descriptions for your key pages.
5. Publish an llms.txt that summarizes your store, and keep it current.
6. Notify search engines when pages change, for example with IndexNow, so fresh content is picked up sooner.