Abstract
AI shopping systems compress product discovery, comparison, and recommendation into one conversation. This changes the role of product data. Attributes are no longer only filters used after a shopper reaches a marketplace. They become inputs to the answer itself. This paper proposes a model that separates product eligibility from recommendation confidence.
Eligibility and preference
Eligibility determines whether a product can enter the candidate set. It depends on identity, availability, price, category, variant structure, and other factual fields. Preference determines whether the product is likely to be recommended for a specific need. It depends on evidence, differentiation, reviews, compatibility, and the quality of the explanation attached to each attribute.
The product evidence graph
A useful product record connects attributes to claims and claims to evidence. A waterproof rating should connect to a test method. A compatibility claim should connect to supported models. A sustainability statement should connect to a standard, certification, or transparent methodology. The stronger the connection, the easier it is to construct a defensible recommendation.
Catalog readiness
Catalog readiness includes stable identifiers, complete variants, normalized categories, accurate inventory, and consistent pricing. Missing or conflicting fields can remove a product from consideration before persuasive content is evaluated.
Recommendation readiness
Recommendation readiness adds use-case language, comparison criteria, limitations, customer evidence, and clear explanations of who the product is for. These signals help the system move from matching a product to defending a choice.
Measurement
AI shopping programs should track product inclusion, recommendation position, attribute accuracy, cited evidence, competitor overlap, and the questions that trigger each product. The most valuable insight is often not whether the product appeared, but which missing signal prevented it from becoming the preferred answer.
Limitations
Shopping experiences differ across platforms, markets, merchants, and data partners. The framework does not assume one universal ranking system. It provides a consistent way to diagnose product visibility across changing environments.
