The outcome nobody reports separately
Most reporting on AI visibility asks one question: did the answer link to your site? It is the wrong granularity. An answer that names your product and links to your home page has done half of the job, and the half it skipped is the one that decides whether the buyer finds the product.
Across 2,185 graded answers between 04.08.2026 and 16.09.2026, where an official URL was present:
| Where the link pointed | Answers | | --- | --- | | the product's own page | 643 | | another page of the same site | 1,445 | | the bare domain, no page | 97 |
Details and method: Do models agree about the same product?.
Why it happens
- The product page is not reachable as a stable URL. Variant parameters,
session tokens and redirects produce addresses a model cannot confidently repeat, so it falls back to something it can: a category, a collection, the home page.
- The page does not say which product it is. Without identifiers in the
HTML — GTIN, MPN, the variant — a model that found the product elsewhere cannot tie that knowledge to your specific page.
- A different page of yours is easier to cite. A comparison or category
page that states facts in sentences is quotable; a product page that hides them in tabs rendered by JavaScript is not.
What you can change
- Give every product one canonical URL, without parameters, that returns the
same page to every visitor.
- Put the identifiers and the key facts in the HTML itself, readable without
running scripts.
- Declare that URL as canonical, and link to it from wherever the product is
described.
What this does not promise
That a model will link the product page next time. We record the outcome, not the reason, and the rules that keep those apart are at /evidence. What can be measured is whether the share moves after a change — which is exactly why the three kinds of destination are worth counting separately.
Sources
- Do models agree about the same product?
- Whose pages AI cites when it answers about a product
- Evidence methodology, version 1.0