The two-minute check
Open a terminal and fetch the page:
curl -s https://your-store.example/products/your-product | lessOr, in a browser, disable JavaScript and reload. Then look for four things in what came back:
- the product name and brand;
- the identifiers — GTIN (barcode), MPN, the variant;
- the facts a buyer asks about — size, capacity, compatibility, what is in the
box;
- a canonical link and the product's structured data.
Whatever is missing there is missing for most crawlers too.
Why storefronts fail it
Modern storefronts often send a nearly empty page and build the product description, specification tabs and variant selector in the browser. A person sees everything; a crawler that does not run scripts sees a title and a price. The specification a model would need to answer "does it fit" is in the part it never received.
What to move into the HTML
- Identity: brand, model, GTIN, MPN, and which variant this page is.
- The deciding facts, written as sentences, not only as a table cell:
"holds a 15-inch laptop", "the 45-litre version, not the 48-litre one sold in 2025".
- Limits: what the product does not do or does not include. Models quote
boundaries as readily as features, and a missing boundary is filled with a guess.
- Structured data — a
Productblock with the same identifiers — as a
second channel, not as a replacement for the visible text.
Why this matters more than it seems
When a model cannot read your page it quotes someone else's. In answers about real products, 37% of citations pointed at the maker's or seller's own site and 44% at press, review and reference pages (study). A review site that states the facts in plain HTML is an easier source than a product page that hides them.
What this does not promise
That the next answer will cite you. Being readable is a precondition, the first steps on the ladder published at /evidence, not an outcome.
Sources
- How AI reads product data
- Whose pages AI cites when it answers about a product
- Evidence methodology, version 1.0