What AI shopping assistants actually read on your Shopify store

Ask ChatGPT for "the best compact stroller for city sidewalks" and you get three or four products, a sentence on each, and a few links. No page two, no ten blue links. If your product isn't in that short list, the shopper never sees it.

That makes it worth knowing exactly what these assistants read, and what makes them pass over a store. The good news: most of it is fixable, and most of the fixes also help your regular Google rankings.

Three kinds of AI visitors

Not every AI "bot" does the same job, and lumping them together is how stores end up blocking the wrong ones.

• Training crawlers collect pages to train future models. GPTBot (OpenAI) and ClaudeBot (Anthropic) are the best known. Google uses a separate robots.txt token, Google-Extended, to let you opt out of Gemini training without leaving Google Search.
• Search crawlers build the index an assistant searches when it answers. OAI-SearchBot (ChatGPT search) and PerplexityBot are the main ones.
• User-triggered fetchers visit a page because a person asked about it right now. ChatGPT-User and Perplexity-User work like this.

You can make a different choice for each. Plenty of brands are fine with search and live fetches but not training. What you almost never want is to block search crawlers or live fetches by accident, and that's exactly what we see most often.

Fix 1: Let them in, for real

A robots.txt that allows AI crawlers is only the first step. The bigger problem is usually further down the line:

• Bot protection switched on years ago, for example a "block AI bots" toggle at your CDN or a security app, quietly refusing every AI visitor.
• Rate limits. Crawlers that ask for many pages quickly get slowed down or refused. We've seen AI crawlers refused on a quarter to a half of product page requests on a store that "allowed" them in robots.txt.
• Slow pages. A crawler on a budget fetches fewer pages from a slow site.

How to check: look at your server or CDN logs for AI user agents and the status codes they got. A 403 or 429 on a product page means that page didn't make it into the answer. Our free AI readiness check tests one product page as an AI crawler and shows what it got back.

Fix 2: Put the facts where a machine can read them

Most AI crawlers don't run JavaScript. If your reviews, your size chart or your spec table are drawn by an app in the shopper's browser, an AI crawler often sees an empty box.

Everything you'd want quoted needs to be in the HTML your server sends: specs, compatibility, dimensions, materials, what's in the box. Then put the same facts into structured data:

• Product with name, description, brand and image
• Offer with price, currency and availability (in stock, out of stock)
• Identifiers: GTIN, MPN or at least SKU
• AggregateRating and individual Review nodes, if you have reviews
• BreadcrumbList, so the engine understands where the product sits in your catalog

Missing price or availability is the single most common gap we find. It costs you in Google Shopping and rich results as well as in AI answers.

Fix 3: Make every fact agree

Assistants compare sources. If your title says "fits 32 to 70 inch TVs", the spec table says 75, the feed says 65 and a retailer listing says 70, the engine has no reason to trust any of them, and a competitor with one clear answer wins.

This is a data problem, not a copywriting problem. The durable fix is to keep each product fact in one structured place in Shopify, in metafields or metaobjects, and have the page, the structured data and the feed all read from it. We wrote about how to do that in Your Shopify store has a single source of truth problem.

Fix 4: Answer the question behind the search

Shoppers ask AI the way they'd ask a knowledgeable friend: "which one fits a small car trunk", "is this safe on drywall", "what's the difference between these two". Pages that answer those questions directly get quoted. Pages that only say "premium quality, best in class" don't.

Good sources for these answers are your own customer service inbox, your reviews and your returns reasons. Put the answers on collection and product pages in plain words, with the numbers up front.

Fix 5: Be talked about elsewhere

Assistants rarely rely on your site alone. Reviews on your product pages, comparison articles, forum threads, retailer listings and videos all count as evidence. You can't fake this, but you can earn it faster: send products to reviewers who compare honestly, answer questions where your customers ask them, and make sure your retailer listings carry the same facts as your own site.

What about llms.txt?

llms.txt is a short Markdown guide to your site for AI agents. It's cheap to add and some agents read it, so we include one in every AI files set. It isn't a ranking factor, though, and it won't fix a blocked crawler or a missing price. Do the five fixes above first.

How to know it's working

Rankings have a position; AI answers don't. The way to measure is to pick a fixed set of the questions your customers ask, run them in each assistant every week, and record who gets named. Over a few weeks you see the trend, and which fixes moved it.

That's what the Command Center does for our clients, across ChatGPT, Perplexity, Gemini, Copilot, Claude and Google's AI Overviews.

A short checklist

1. AI search crawlers and live fetchers allowed in robots.txt, and not refused by your CDN or apps.
2. Product pages load fast and return 200 to AI user agents.
3. Specs, compatibility and reviews present in the server-sent HTML.
4. Product structured data with price, currency, availability, brand and an identifier.
5. One source for each product fact, used by the page, the schema and the feed.
6. Collection and product pages that answer the real questions in plain words.
7. A weekly check of who AI names for your key questions.

Run the free AI readiness check to see where your store stands on the first four.