AI mention

Count a mention only when the entity is visibly present in the preserved answer or its structured result. Do not turn that occurrence into an intent, rank, or quality judgment that the response does not express.

A mention records presence; it does not imply that the entity was linked, cited, recommended, or used as the answer’s evidence. Code each of those events independently, even when they appear next to one another.

Capture the exact wording, entity span, query, surface, locale, region, timestamp, account state, and any accompanying link. Note ambiguous names and aliases so that a string match is not mistaken for a confirmed entity mention.

Test the mention rule on positive, negative, and ambiguous examples before coding the full sample. Keep the original answer beside every label, because the surrounding sentence often determines whether a name refers to the intended entity.

A mention rate answers how often an entity appeared in the sampled answers. It does not measure prominence, sentiment, traffic, attribution, or conversion unless those outcomes are separately observed and reported.

Mentions are sensitive to spelling, aliases, language, and the boundary of the answer being coded. A sampled occurrence may miss implied references or falsely include a namesake, so entity resolution and uncertainty must remain visible.