AI assistants recommend products from the structured data on retailer product pages — titles, descriptions, ingredients, specifications and schema markup — cross-referenced between retailers. When that data is patchy or contradicts itself from one listing to the next, assistants quietly skip the product. Consistent, machine-readable data is what keeps a brand in the recommendation set.
This is not a forecast. It is measured traffic, and the curve is steep.
Source: Adobe Analytics, 2025.
Adobe's research carried a second finding that matters more for brands than the headline growth: retail sites lag in AI-search visibility because their content is not machine-readable. Shoppers are arriving via assistants faster than product pages are becoming legible to those assistants. That gap is the AI shelf — and a product is either on it or it is not.
An assistant does not browse the way a shopper does. It reads the structured layer of a product page — title, description, ingredient and composition list, feeding guidance, specifications, pack sizes, schema markup — and then it does the thing humans rarely bother to do: it cross-references the same product across every source it can reach.
That cross-referencing step is where products fall off the AI shelf. Two different ingredient lists for the same recipe. Three different feeding guides for the same bag. A protein percentage that changes depending on which retailer you ask. To a human these are shrugged off as retailer quirks; to an assistant they read as unreliability — and an unreliable product is easier to leave out of an answer than to explain.
This is not a rare edge case. 96.5% of matched products have an ingredients issue — the list missing or mismatched somewhere they're sold — across UK pet retail.
Eagre analysis of 6,000+ EAN-matched products across 14 UK pet retailers, July 2026.
The uncomfortable implication: a brand can have excellent data in its own systems and still look contradictory in public, because the versions that assistants actually read live on major UK pet retailers' pages — not in the brand's spreadsheet.
What assistants read this quarter is not what they will read next quarter. Models change, schema conventions change, answer-engine optimisation changes with them.
A brand managing a dozen retailer templates by hand cannot chase a moving target. Every AEO change becomes a dozen edits, a dozen email threads and a dozen chances to reintroduce a contradiction.
With a single approved golden record per product, an AEO-driven change is made once and adapts every connected listing at sync speed — with the retailer keeping the go-live gate on their store.
The point is not “optimised today”. It is being structurally able to stay optimised as the target moves — readiness and consistency, not a promise of placement.
Pet owners increasingly ask assistants about food and treatment choices — which diet for a sensitive stomach, whether an ingredient is safe alongside a medication, how much to feed a senior cat. Health-adjacent answers are held to a higher bar: assistants are noticeably more conservative about recommending products when the underlying data is incomplete or inconsistent.
That caution rewards complete, consistent data even more. In the categories where UK specialist pureplayers do their most valuable trade — diets, supplements, parasite treatments — the products with one clean, agreeing story across every listing are the ones an assistant can safely put in an answer. Everything else gets the quiet skip.
Assistants recommend what they can read and trust. One golden record, consistent on every connected listing, keeps your products legible as the AI shelf keeps moving.