If you have ever typed your brand into ChatGPT, you probably came away with a view of how AI models see your store, and it may have shaped what you have done since. That view rests on a single answer. To find out which of your pages could have answered the question, read the store itself, field by field.
Take each product’s sheet from chapter 3 and read every field and page that describes the product against it. Places that disagree go on a list called Fix first, true facts missing from the text go on Add, and pages that are already right go on Keep.
The AI models in this guide are ChatGPT, Gemini, Perplexity, Claude and Google’s AI Overviews. 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. This chapter follows the Overnighter 30, a backpack for work trips of one to three nights.
What does Shopify already send to AI models?
If your store is eligible, Shopify already makes your products available to ChatGPT, Gemini and Google’s AI Mode, Microsoft Copilot and Meta’s AI. It does this by default. What your product fields say is what those AI models receive from Shopify.
Shopify calls this Agentic Storefronts (opens in a new tab). It is the connection that puts your products inside AI models’ shopping features, where shoppers can find them and buy them. The products travel through Shopify Catalog (opens in a new tab), a structured copy of your product information that Shopify keeps and shares with those AI channels. You can stop any one channel from receiving it, and Shopify Catalog itself stays on.
You will find it in your admin under Sales channels, then Agentic. Since 11 May 2026 (opens in a new tab) it has its own page, which shows how your products do in each AI channel and suggests improvements to your product data.
Shopify’s help pages say it shares titles, descriptions, images, categories, variants, options, prices and availability, and that each AI channel decides the final ranking and wording. OpenAI says (opens in a new tab) Shopify merchants need to do nothing extra for their products to reach ChatGPT.
Being available puts a product among those an answer can name. Whether an answer names it depends on what your pages and other sites say about it, which How AI search decides which brands to recommend explains. The rest of this chapter is about your side of that.
Which parts of a product does an AI model read?
It reads the fields, and a product has many more of them than its description. A field is one box in your admin that holds one piece of a product’s information. Open any product in your Shopify admin and you will find these.
- The title and the description.
- The product type and the category.
- The vendor, and the tags.
- Options and variants, such as colours and sizes.
- Image alt text, the short description of each picture that screen readers read aloud.
- The search engine listing, Shopify’s name for the title and description shown in search results.
- Metafields, the extra fields your theme or apps add, such as a features list or a materials box. A metafield appears on the page only if your theme is set up to show it (opens in a new tab).
Other pages describe the product too. The collection it sits in has a description of its own. Your FAQ page may answer questions about it, and so may a product FAQ, the block of questions and answers some themes show on the product page, and old blog posts may mention it by name. Each of these is part of what your store says about the product.
These fields reach AI models in two ways. Whatever your theme prints on the product page reaches any AI model that opens the page. The fields Shopify names travel through Shopify Catalog to the AI channels that use it. Both start from the same boxes in your admin, which is why reading a product means reading all of them.
Do you need llms.txt, a schema app or special markup?
Google says (opens in a new tab) you need no special file, markup or Markdown for its AI features. The work that counts is the words in your fields.
You will meet two things in app listings and sales pitches. One is llms.txt, a text file placed on a website to give AI tools a summary of its pages, which was proposed in September 2024 (opens in a new tab) and is still a proposal. The other is structured data (opens in a new tab), code in a page that labels its contents for search engines, such as which number is the price. Shopify apps sell both.
Google’s guide says you don’t need machine-readable files, AI text files, markup or Markdown to appear in Google Search, and that Google Search ignores them. It says structured data is not required for its AI features either. Google still recommends it for the rest of your search work, where it can earn the richer listings Google shows for some products, such as a price under the link.
OpenAI says (opens in a new tab) ChatGPT weighs product data from providers such as Shopify, which comes from your fields, together with what other sites say. To see what structured data a product page already carries, paste its address into Google’s Rich Results Test (opens in a new tab).
How do you check one product?
Go down the product’s sheet, and for each reason ask three questions. Start with the products your chapter 2 questions are about.
- Is it written as text? A fact that lives only in a photo, or in a tab that loads after the page, counts as missing. Chapter 1’s page-source check takes about a minute.
- Do your own pages agree? Read every place on your store where the fact appears, from the product page to the FAQ, the collection and old blog posts, and read the listings you send to retailers and marketplaces too, because you wrote them. A granola maker that switched to a different oil may find its own product page right and its retailers’ listings still naming the old one.
- Where else is it said? Search for the product by name and note the outside pages that state the fact. Leave them for now. They go on chapter 6’s list.
Write each finding as a plain statement beside the document, such as The FAQ says the harness is fixed. The 2025 design spec says the straps stow.
Anyone can check a finding written this way, and it tells the next person exactly what to change.
What usually turns up?
Expect two things on a first read. The products that sell best often have the least written about them, because they get repriced and restocked again and again while the careful copy goes to whatever launched last.
And a fact sometimes lives in one odd place only, such as a tag, a search description someone filled in years ago, or a PDF. A desk lamp’s product page might say dimmable
while its four dimming modes are described only in the manual you download. Some of those places vanish in a rewrite and some sit where few shoppers look. Either way, the fact goes on the Add list.
What if your own pages disagree?
The places where your own pages disagree come first. An AI model that reads two pages saying different things about one fact has to pick one, and it may pick the one that is out of date.
Tarrowby redesigned the Overnighter 30 in 2025 so that the shoulder straps stow behind the back panel. The product page was updated that week. The FAQ still says the harness is fixed. Tarrowby’s own Amazon listing says padded, fixed shoulder straps, and Outside’s 2024 review says the straps can’t be tucked away. A YouTube review from 2025 has it right.
The 2025 design spec, ranked first in chapter 3’s order, settles which is right. The FAQ is Tarrowby’s own page, so it goes on the first of the three lists below. The Amazon listing and Outside’s review sit on other sites, so they go on chapter 6’s list, even though Tarrowby wrote one of them.
What do you fix first?
Fix the places that disagree first. Sort everything you found into three lists, Fix first, Add and Keep, and finish the first list before you start the second.
Fix first holds every place where two of your own pages disagree about one fact. It comes first because chapter 5 copies each fact into more fields, and copying a fact that is still unsettled spreads the wrong version further.
Add holds what is true, proved by a document and missing from the text of your pages. Put it in order by how often chapter 2’s questions ask for each fact. If more of Tarrowby’s questions are about getting a laptop out at airport security than about a daypack that folds into its own pocket, the laptop zip goes first.
Keep holds the pages that already state their facts plainly. Write down the date you checked each one. They show what the others should look like, and they tell whoever does the rewriting what to leave alone.
Anything you found on another site goes on chapter 6’s list instead.
There is always more to fix than time to fix it. Indaeo AI ranks each problem it finds by the evidence behind it, so you start with the ones that matter most. When an AI answer gets a fact about your product wrong, Indaeo AI traces it to the page the answer took it from.
Chapter 5 is the writing, done in the order of these lists.