The product is an entity, and almost nobody structures it as one
A search engine matching a query to a page can work with a page. An assistant assembling an answer about a product needs to know what the product is, its identity, its attributes, how it relates to other products, and whether the thing described on your site is the same thing described on three marketplaces and a review site.
That is an entity problem, and it sits one level below the brand entity that service businesses spend all their effort on. Most catalogs have never had it addressed, which is why so much ecommerce AI-visibility work produces nothing: the brand gets tidied and the products stay illegible.
What actually gets built
Product and Offer schema that is complete rather than present. Most catalogs emit Product markup with a name and a price and stop. The fields that determine whether an assistant can use the product, identifiers, condition, availability, shipping and return detail, aggregate rating where it is real, are the ones most often left empty.
GTINs and stable identifiers wherever they exist. The single most reliable way for a machine to confirm that your listing and a marketplace listing describe one product. For one-of-a-kind inventory where no GTIN exists, the work is different: internally consistent identifiers plus descriptive attributes precise enough to disambiguate.
A taxonomy that matches how buyers ask. Catalog structure usually mirrors how a merchant thinks about stock. Buyers arrive with a use, a constraint or a comparison, and the categories that answer those questions frequently do not exist as pages at all.
Attribute coverage against what assistants actually read. Structured product data has expanded considerably in what it exposes to AI surfaces, and the attributes that matter are not always the ones a merchandising team would prioritise.
The one-of-a-kind problem
Catalogs of unique inventory, antiques, vintage, art, salvage, bespoke, break most standard ecommerce SEO advice, because the advice assumes a product that persists. When a piece sells and never returns, a page built to rank for that piece is a page that decays to nothing.
The work here shifts up a level: the durable asset is the category, written to be definitive about the type of object rather than about one instance of it. Individual pieces illustrate; the category page is what accrues authority and gets cited. Copy has to be written so it stays true as stock turns over, describing what a collection has included rather than what is in it today.
Fair questions
We already have Product schema. Isn't that done?
Usually not. Emitting Product markup and emitting the fields that make a product usable to an assistant are different things. The audit checks completeness against what the surfaces actually read, not whether the markup validates.
Does this help traditional SEO too?
Yes. Better product entities improve rich results and merchant surfaces independently of anything AI-related. It is one of the few areas where the AI-era work and the classic work are the same work.
What if our products have no GTINs?
Then identity has to be established differently, consistent internal identifiers plus attribute precision. It is harder and it is the normal case for unique inventory. It is not a blocker.
See where your catalog stands
A free Visibility Check covers product-entity legibility alongside the rest of the six-layer audit.
Get my free Visibility Check