Why ChatGPT and Gemini Don't Recommend Your Store

ChatGPT and Gemini skip your store for one of three reasons. Your product pages lack the schema.org markup a model needs to extract price, availability and identity with confidence. Your product feed is missing fields like GTIN, brand or condition. Or you have no structured review data for the agent to cite as social proof. Fix those three things and most stores start showing up within weeks.

The three reasons AI agents skip stores that rank fine for humans

Being indexed and being recommended are two different jobs. Google or Bing can crawl your product page, list it in search results, and still hand nothing usable to an AI agent trying to answer "find me a waterproof jacket under $150." The agent isn't reading your page the way a shopper does. It's extracting structured facts, and if those facts aren't marked up cleanly, it moves to a competitor's page instead.

In the audits we run at UXFix, the pattern repeats across categories. A store has clean HTML, fast load times, correct on-page pricing, and still gets zero mentions when we test the same catalog through ChatGPT and Gemini shopping queries. The cause is almost never crawlability at that point, we cover that gap in Why AI Shopping Agents Can't Find Your Product Pages. The cause is that the page gives a human eye enough context to buy, but gives a parser nothing to hang a recommendation on.

Three gaps show up over and over. First, no schema markup at all, or schema that's present but broken enough that validators reject it. Second, a product feed built for a shopping ad campaign years ago that was never updated with GTIN, brand or current stock status. Third, reviews that exist visually on the page as stars and text, but aren't wrapped in any markup a model can parse as a rating.

What an AI-recommendable product page actually looks like

An AI-recommendable product page is one where every fact an agent needs to complete a purchase decision, name, price, currency, availability, brand, identifier and rating, exists in structured form, not just as visible text. A shopper can infer "in stock" from a green checkmark. A model needs "availability": "https://schema.org/InStock" in the markup to say the same thing with confidence.

Use this as a working checklist before you touch anything else:

• Valid Product schema on every product page, including name, image, description, sku and brand.
• Offer schema nested inside it with price, priceCurrency and availability kept in sync with what's actually in stock.
• AggregateRating or Review schema if you display reviews anywhere on the page. Visible stars with no markup are invisible to an agent.
• A product feed (Google Merchant Center or equivalent) with GTIN, MPN or brand populated for every SKU, not just your bestsellers.
• Consistent price across page, schema and feed. A mismatch is one of the most common triggers for an agent to drop a listing entirely, covered in more detail in Why AI Shopping Agents Misread Your Prices.

Run any product URL through Google's Rich Results Test or the Schema Markup Validator before assuming it's fine. We see stores that added schema two years ago, then redesigned the page and silently broke the JSON-LD block. It still renders fine for a shopper. It returns nothing for a parser.

Schema markup AI shopping agents actually read

Not all schema is equal weight. Agents built on top of Google's Shopping Graph and general LLM retrieval both lean hardest on a small set of schema.org types, and the rest is nice to have.

The schema.org Product type and Google's own structured data documentation are the two references worth bookmarking. Both are updated as agent behavior shifts, and both are free to validate against without a developer.

One detail that trips up otherwise careful teams: dynamically injected schema. If your JSON-LD block is added by a client-side JavaScript tag manager after the page loads, some crawlers see it and some don't. Server-rendered schema, present in the initial HTML response, is the safer bet for any agent that doesn't execute JavaScript before extracting facts.

Why doesn't Gemini show my products even when Google indexes them?

This confuses a lot of store owners, and reasonably so. Gemini's shopping features and Google's Shopping Graph pull primarily from your Merchant Center feed, not from the organic index that decides whether your page ranks in a normal search. You can be perfectly indexed and completely absent from that graph if your feed is thin, outdated or missing required attributes.

The feed requirements for AI shopping agents are stricter than what most stores set up for basic shopping ads. Availability has to update in near real time, not overnight. GTIN is effectively mandatory now for most categories, not optional. Price in the feed has to match price on the page down to the cent, including whether tax or shipping is baked in.

If you sell through a platform like Shopify or BigCommerce, the native feed apps get you most of the way there, but they don't fix missing GTIN or bad category mapping automatically. Someone still has to check the actual feed file, not just assume the integration handles it. We also see checkout-stage issues compound this: an agent that likes your product but can't complete a purchase because of blockers later in the funnel will still get scored down, which we detail in Why AI Shopping Agents Abandon Checkout on Your Store.

How to get AI shopping agents to recommend your products

Treat this as a three-part audit, not a single fix. Schema, feed, and reviews each need separate attention because they fail independently and get checked by different systems.

Start with the highest-traffic 20 SKUs and get those fully compliant first. That gives you a fast read on whether the fixes move the needle before you roll changes across a full catalog of a few thousand products. NN/g's research on how people scan and trust structured content is a useful lens here too, the same clarity that helps a human shopper decide fast is what a model needs to extract fast, see NN/g's usability research for the human side of that argument.

Frequently asked questions