Open ChatGPT and type "best noise-cancelling headphones under €150." You’ll get a confident, specific answer: a few named products, with reasons. Now ask Perplexity or look at a Google AI Overview for the same thing. Same pattern. None of them hand you ten blue links. They hand you a recommendation.
Here’s the uncomfortable question for any online store: when AI answers a question your product could have answered, does it mention you, or your competitor?
For most stores right now, the honest answer is "neither, and I have no idea." That’s the problem worth understanding.
Shopping stopped being a search box
For twenty years, getting found meant ranking on a results page. A shopper typed a query, scanned links, and clicked. Your job was to be link number one.
That behavior is splitting in two. People still search, but a fast-growing share now ask. They ask an assistant to compare, to recommend, to shortlist. And an assistant doesn’t show ten options for the shopper to evaluate. It picks. The shortlist is the result.
If you’re not on the shortlist, you don’t get a worse ranking. You get nothing. There’s no page two to crawl back from.
Why AI can’t see most product catalogs
AI assistants don’t browse your store the way a human does. They rely on structured, machine-readable data: clean product information in formats they can parse with confidence, like schema.org / JSON-LD markup and the product feeds that flow into shopping and AI surfaces.
Most catalogs fail that test in boring, fixable ways:
- Missing or thin structured data. No JSON-LD, or partial markup that leaves out price, availability, GTIN, or brand. The assistant can’t trust what it can’t verify, so it skips you.
- Titles written for humans, not machines. "Our bestseller, now back in stock!" tells an AI nothing. "Sony WH-1000XM5 Wireless Noise-Cancelling Headphones, Black" tells it everything.
- Inconsistent attributes. Color in one field, sometimes in the title, sometimes nowhere. Size missing on half the variants. The assistant needs reliable attributes to match a query like "black, over-ear, under €150."
- A feed built only for Google Shopping. Even a clean Google Shopping feed isn’t automatically readable by ChatGPT or the newer AI-shopping formats. They overlap, but they’re not the same.
None of this shows up as an error. Your site works. Your ads run. You just quietly never appear in the answer.
How to check if you’re invisible (a 5-minute self-audit)
You don’t need a tool to get a first read:
- Ask the assistants directly. In ChatGPT and Perplexity, ask the exact question a buyer would ask in your category ("best [your product type] for [use case] under [price]"). Do you appear? Do your competitors?
- Check a Google AI Overview. Search your main category query and see what the AI Overview cites. Is it your type of store, or marketplaces and review sites?
- View one product’s source. Right-click a product page, "View Source," and search for
application/ld+json. If there’s no Product schema, or it’s missing price/availability/GTIN, AI is working with less than it needs. - Open your product feed. Look at ten titles. Would a stranger, or a machine, know exactly what each product is from the title alone? Are color, size, and brand present and consistent?
If those checks make you wince, you’re not behind because your products are worse. You’re behind because your data is harder to read than your competitor’s.
The fix: make the catalog readable, then keep it that way
The good news: this is a data problem, and data problems are fixable. Getting AI-visible comes down to three moves.
- Clean the structured data. Complete, valid Product JSON-LD on every page (price, availability, GTIN, brand, attributes), and a feed that carries the same clean data into shopping and AI surfaces.
- Write titles and attributes for machines and humans at once. Specific, attribute-rich titles. Color, size, material, and category filled in consistently across the whole catalog, not just the hero products.
- Publish in the formats AI actually reads. Beyond the classic Google Shopping feed, that increasingly means structured JSON-LD and the emerging AI-shopping feed formats, so ChatGPT, Perplexity, and AI Overviews can all consume the same clean catalog.
Do this once and you stop being invisible. Keep it clean as your catalog changes, and you stay visible while competitors drift back into the dark.
What you can measure, and what you cannot
This is where most advice on AI visibility goes quiet, so here is the uncomfortable version.
You cannot currently isolate traffic from AI answers in your analytics. A visit that started with an AI assistant recommending you mostly arrives looking like direct traffic, or like an organic click, or like nothing at all when the assistant answered the question and the shopper never clicked. There is no reliable channel to filter on, and any tool promising you a clean number for it is estimating. Treat a confident figure here as a claim about the vendor, not about your store.
What you can measure is whether your data is readable, and that turns out to be the more actionable half. Completeness per required field, how many products carry identifiers, how many have the attributes people filter on, how many are disapproved in Merchant Center for a data reason. None of these tell you an assistant recommended you. All of them tell you whether it could.
The second thing worth watching is your branded search volume. If assistants are surfacing you, more people go on to search your name, and that shows up in Search Console where you can actually see it. It is indirect and it lags, and it is still the most honest signal available.
The practical consequence is a change of question. “How much traffic are we getting from AI?” has no trustworthy answer yet. “Can an assistant answer a question about our products from our data?” has a precise one, and you can improve it deliberately. That is the number a feed grade gives you.
Two neighbouring pieces if you want the mechanics rather than the strategy: the disapproval reasons Merchant Center reports are the same data gaps in a system that does tell you about them, and the OpenAI ACP product feed is what the requirements look like when an agent, not a shopper, is the reader.
This is exactly what Verintra does
We built Verintra for this shift. It scans your feed, grades every product A→F on how machine-readable it is, uses AI to fix the titles, descriptions, and missing attributes holding products back, and publishes the result in the structured formats that Google Shopping and AI assistants read. You see, in plain language, what’s making you invisible and what to fix first.
The first step is free, and it answers the question this whole post is about: scan your feed and see exactly how AI-ready your catalog is right now.
Scan your feed free at verintra.com. No credit card, no contract. Two minutes to find out whether AI can see your products at all.
