When somebody asks an assistant which brand to buy from, the answer names two or three and moves on. Your product page is almost never what gets quoted. What gets quoted is a comparison article, a buying guide, a forum thread, or a page on your site that explains something rather than sells something. So the job for a store is to be present in the consideration, not to rank a product URL.
That single difference is why GEO advice written for software brands does not port cleanly. A software company can publish a documentation page and be quoted from it. A store selling physical goods has to earn its way into a recommendation somebody else is making.
The pages a store can actually get quoted from
Sizing and fit, materials and care, comparisons between your own products, and the honest answer to the question people ask before buying your category. These are the pages assistants reach for, because they contain a complete answer that does not depend on being logged in or on a price that changes.
What almost never gets quoted: a collection page, a page whose content is a product grid, and a product description written as ad copy. None of them contain a passage that answers a question.
The assistant is deciding who gets considered. The product page is what happens after that decision, not what causes it.
The catalogue still matters, for a different reason
Structured product data does not get you cited in a conversational answer, but it is what makes you visible in the shopping surfaces attached to those answers, and it is the same data your Shopping campaigns depend on. Get the identifiers, the attributes and the availability right once and both benefit. Getting it wrong is visible in Merchant Center long before it is visible anywhere else.
Third party surfaces are most of the work
For a store, the sources an assistant trusts about your category are usually not yours. Review platforms, reddit threads, buying guides in trade press, and the roundup articles that already rank for your category. Being named in three of those does more than another page on your own domain, and it is slower work with no dashboard attached to it, which is why it is so often left out of a GEO proposal.
What it looked like on our own store
Phrase Studios is a clothing label we owned. Over six weeks the work was rewriting existing pages, then publishing a Journal on top of them, then adding an llms.txt. AI feature impressions went from 290 to 23 503 and Copilot citations went from 219 in July to 8 257 in August. Search went from 13% of sessions to 45%. The base was small, we say so, and the case study carries the exports rather than a summary screenshot.
How to measure it as a store
Ask the assistants the five questions a customer would ask before buying your category and record which brands get named, monthly, in the same document. Then watch Search Console's AI mode filter and Bing Webmaster Tools for the Copilot side. Referral traffic from assistant domains will be small and is not the point: the point is whether you are in the consideration set.
What we do with it
We start with whether the crawlers can reach you, then the pages that could be quoted, then the third party surfaces where your category is decided. It runs alongside the store work rather than instead of it, and both are on the SEO and GEO page.