GEO: how ChatGPT decides which stores to recommend

How does ChatGPT choose which products to show?

OpenAI explains it in its help article on shopping with ChatGPT search. When a question shows shopping intent, ChatGPT can show products with images, details, and links to sites where you can buy them. A product appears when ChatGPT judges it relevant to the shopper's intent, based on the question and on context such as Memory or custom instructions.

“

Product results are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships.

OpenAI Help Center, “Shopping with ChatGPT Search”

To decide which products to show, OpenAI says ChatGPT considers three things:

• Structured product data from first-party and third-party providers, like the price and the product description, and other third-party content.
• ChatGPT's own first answer, the response it generates before it looks at any new search results.
• OpenAI's safety standards and product policies.

A few details matter for a store. ChatGPT may write its own simpler titles and descriptions from the data it receives. Labels like “Budget-friendly” are generated by the model and aren't verified. Review summaries come from reviews on public websites. And prices come from third-party providers, so a price change can take a while to show up.

How does ChatGPT rank stores that sell the same product?

When a shopper opens a product, ChatGPT may list the stores that offer it. OpenAI says the list is built from merchant and product data, supplied by third-party providers or by the merchants themselves. The ranking factors it names are availability, price, quality, and whether the store is the maker or the primary seller of the item. OpenAI adds that it expects to personalize these rankings more over time.

How your product data gets there depends on your platform. For stores on Shopify, OpenAI says the product data is already connected to ChatGPT through Shopify Catalog, with no extra work. Other merchants can read OpenAI's product feed documentation and apply to send a direct product feed.

Can AI tools read your site at all?

Before any of this, an engine has to be allowed to read your site. Each company runs separate crawlers for search and for training, and you control each one in your robots.txt file:

The difference matters. OpenAI says each setting is independent: you can keep GPTBot from training on your content and still appear in ChatGPT search, as long as OAI-SearchBot is allowed. A site that blocks OAI-SearchBot isn't shown in ChatGPT search answers, though it can still appear as a navigational link, and a change takes about 24 hours to apply.

Anthropic says blocking Claude-SearchBot may reduce your site's visibility and accuracy in its users' search results. Perplexity recommends allowing PerplexityBot if you want to appear in its results. And Google says Google-Extended doesn't affect whether your site appears in Google Search, and isn't used as a ranking signal.

What does Google say about AI Overviews?

“

There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.

Google Search Central, “AI features and your website,” updated Dec 10, 2025

To be shown as a supporting link, a page must be indexed and eligible to appear in Google Search with a snippet. That's all. Google adds that meeting the requirements doesn't guarantee a page will be crawled, indexed, or shown.

The basics it lists are the familiar ones: let crawlers in, link your pages to each other, put important content in text, back it with good images and video, make your structured data match the visible text, and keep your Merchant Center and Business Profile information current. It also says you don't need new machine-readable files, AI text files, or special markup to appear in these features. In other words, at Google, GEO is mostly good SEO.

What did the GEO research find?

The term GEO, short for generative engine optimization, comes from a research paper first posted in November 2023 and accepted at the KDD 2024 conference. The researchers built a test set of 10,000 queries and tried 9 ways of rewriting web content, to see what made a source more visible in AI-generated answers.

The methods that worked best were plain ones: citing sources, adding quotations from relevant sources, and adding statistics. The researchers also report that pages ranked lower in classic search gained the most from these changes, while keyword stuffing, an old SEO habit, didn't help.

Two cautions. Most of the testing ran on the researchers' own setup, built on GPT-3.5, an older OpenAI model, and only part of it on a live engine, Perplexity. And AI engines change all the time, so what worked in 2024 may not work the same way now. Treat it as a direction, not a formula. It's a sensible direction anyway: content with sources and numbers is better for your readers, too.

Does llms.txt help?

llms.txt is a proposal by Jeremy Howard, first published on Sep 3, 2024 and updated to a second version on Aug 10, 2026: a plain-text file at your site's root that tells AI tools what's on the site and where to find it. Its own page says thousands of sites now publish one, and that Chrome's Lighthouse checks for one in its agentic browsing audits.

It's still a proposal, not a standard. Google says you don't need AI text files to appear in its AI features, and none of the crawler pages from OpenAI, Anthropic, or Perplexity says its crawler uses llms.txt to decide what to show. It's cheap to publish, so treat it as a small extra next to readable pages and accurate structured data.

What can you do this week?

1. 1Check your robots.txt. Open your site's address with /robots.txt at the end, and make sure OAI-SearchBot, Claude-SearchBot, PerplexityBot, and Googlebot aren't blocked. Then check your host's bot settings.
2. 2Make sure your content is in the page. Open a product page's source, not the inspector, and look for its name, price, and description. That's what a crawler that doesn't run code reads.
3. 3Make your structured data match the page. Price, availability, shipping, and returns, exactly as the page shows them. Google's Rich Results Test lists what it finds.
4. 4Keep your product facts the same everywhere. OpenAI ranks stores by availability, price, quality, and whether they make or primarily sell the item.
5. 5Write pages that answer real questions, with sources and numbers. In the GEO research, that's what made content more visible.
6. 6Ask the engines yourself. Ask ChatGPT, Perplexity, and Gemini the questions your customers ask, and see which stores are cited, and from which pages.

What does MOVA do for AI search?

Every MOVA site is built to the usual guidance for Google and AI search. Here's what's in place, and how far each part goes:

• Ready-made pages and llms.txt. For a store moving to us from WordPress, our pipeline builds ready-made pages an AI tool can read without running code, and an llms.txt file that says who you are and what you sell.
• AI crawlers welcome. At Hasod, the online gift store in Israel where we built MOVA, the robots file welcomes GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot, and allows Google-Extended, the setting that lets Google use the site's pages for Gemini.
• Structured data on product pages. At Hasod, product pages add their price, shipping details, and return policy as structured data once the page loads. Google Search reads it when it renders the page. A crawler that doesn't run code won't see it, and Google notes that product markup added with code can make its Shopping crawls less frequent and less reliable.
• Titles and descriptions in one place. At Hasod, the title and description Google shows for each product and category page are written in one place, in the admin.

A site isn't built once and forgotten. Once Lumi, your AI website agent, joins your store, she checks it on a schedule. At Basic, that means a site check once a month: speed, broken links, and titles for Google. The team joins one department at a time.