Products: your feed and product pages, as AI reads them

What it does

AI answers and AI shopping results recommend products from what they can read: a shop's product feed and the product pages themselves. The Products page (under AI visibility) imports your catalog, reads each product's page the way AI crawlers do (the HTML, without running JavaScript) and flags what to fix, next to how the AI answers you track use each product. It costs no credits.

Importing your products

Managers and admins import and delete the catalog and read the pages; everyone who can see the project can see the page and download the CSV.

• Feed or store address: your Google Merchant Center feed (XML as RSS or Atom, or tab- or comma-separated with a header row of attribute names; gzip too), or your store's address: Shopify and WooCommerce stores are read from their public product list. The import runs in the background and the page refreshes by itself.
• Feed file: the same formats as an upload, up to 10 MB (gzip a bigger file).
• Product pages: no feed? Paste product page addresses, up to 500, or take the pages the AI answers you track cite or read and the pages of your latest Google AI report. Pages that turn out not to be product pages are set aside.
• Up to 5,000 products per project. From a bigger feed, the products that aren't out of stock come first, then out-of-stock ones fill the rest, in the feed's order; a store is read up to 5,000. A feed is read up to 60 MB (unzipped), and a feed that stops arriving (too big or too slow) is read up to where it stopped. The page says how many products the source has at least.
• Importing again replaces the catalog; a product that keeps its id and page keeps what its page check found. If an import fails, the previous catalog stays.

Reading the product pages

Read the product pages reads up to 250 pages per check, one at a time so your server isn't rushed: pages never read come first, and among them the products the AI answers cite, name or read. Run it again for the next ones. Only pages on the project's own sites are read: its domains, the brand's website and the store's address. Products that share a page share one read.

What gets flagged

• Page doesn't open: not found, an error, or no answer. A product link that leads nowhere is also what shoppers and AI answers reach.
• No product structured data: the HTML has no schema.org Product (JSON-LD, microdata or RDFa), or its product JSON-LD isn't valid JSON, which search engines skip too.
• Product data only after JavaScript: the HTML has none, but the page's scripts carry it (a Next.js or Nuxt app, a store's product script, a script that writes the JSON-LD). Most AI crawlers don't run JavaScript, so they see none of it.
• No price in the structured data: no offer with a price, so nothing to quote.
• Price or stock mismatch: the feed or store says one thing and the page's structured data another. For a product with variants, the page's variant with the same SKU or GTIN is compared; a price inside the page's price range counts as the same.
• AI quotes another price: an AI answer names the product next to a price that isn't its price (see below).
• No GTIN (EAN, UPC or ISBN), or one whose check digit is wrong. A product that has none can say so in the feed (identifier_exists: no).
• No SKU or MPN in the structured data, so readers can't tell this exact product from similar ones.
• Variants merged into one: the product comes in variants (sizes, colours) but its structured data describes one product, with one offer or a price range; or the feed has one row for variants the page lists.
• Short description: under 50 words.
• Weak title: under three words, codes only, in capitals, without the brand, or over 150 characters (shopping results cut it).
• Photos without alt text: the product's photos in the page's HTML with no alt text.

What the AI answers do with each product

Each product shows how the AI answers tracked for the project in the last 30 days use it: the answers that cite its page as a source or read it, and whether the page is in your latest Google AI report. Named counts the latest answer of each prompt on each engine that names the product by the words of its title up to its model, such as “Galaxy S24”, “WH-1000XM5” or “Silk-épil 5-251” (a model with a hyphen is one model). A model's words must stand together and in order, a number followed by a quantity (“30 kosov”) isn't a model, and a name of generic words and a number (“Premium 30”) also needs the brand nearby; a title without a model needs the brand nearby too, and plain titles such as “Oak table” aren't matched.

A title without a model, or with a single digit, names a product line rather than one product: “Pampers Premium Care”, “Bosch Serie 6”, or a tyre without its size (“PremiumContact 7”: the size, load and speed index and DOT code are left out). Answers naming the line show as “its line, … named”, and the prices they quote aren't compared, because they may be for another size or model.

When such an answer gives a price next to the name, in the product's currency, it is compared with the product's price: more than 5% off (and at least 1) is flagged, with the answer's words. Products named alike, such as two memory sizes of one phone, share their mentions, and a price that is either one's counts as right. Budgets (“under 300 €”) aren't read as prices.

Google Shopping

ChatGPT fills its product carousels from Google Shopping: in a 2026 study, 83% of the products in its carousels were among Google Shopping's top 40 organic results for the carousel's own searches, and 60% among the top 10. The Google Shopping card shows where your store's listings rank for the searches you choose (up to 25 per project).

• Add searches from your tracked keywords (with their own country and language), from the searches the AI engines ran for your prompts that ask for products (ChatGPT's first), or type your own. ChatGPT's carousel runs its own, shorter searches, which it doesn't show; these are the closest you can check.
• A check reads the top 40 organic listings; sponsored ones don't count. A listing is yours when its site is one of the project's domains or its website, or its seller is your brand. Your listings are matched to your catalog's products by title, and a matched product shows its best place in the list of products.
• Check now is charged per search (the button shows the credits). Every Monday the searches are checked again, charged the same way, when the project checks anything on a schedule (rankings or AI visibility). A check that fails is refunded.
• A search can show "Not in the top 40": Google Shopping lists other stores' offers, or, for a product many stores sell, one listing for them all.

Good to know

• Pages are read like a visitor's browser asks for them. Pages behind a login or a bot wall show as not opening.
• Prices are compared in the same currency only.
• The CSV lists every product with its flags and what each one found; pick a flag first to download just those.
• Deleting the catalog deletes its products and what the page check found.