A brand mention is any time an AI answer names a brand. It is the atomic unit of AI visibility, from which ranks, scores, and share of voice are computed.
Not all mentions are equal. Position matters (first named beats sixth), framing matters (recommended beats listed beats warned-against), and consistency matters (named every time beats named once). Scoring systems weight all three.
Mentions are also where measurement must stay honest: a mention only counts if it appears in a genuine recorded model answer. An index that cannot show you the answer behind a mention is asking you to take its numbers on faith.
What counts, and what does not
A mention is a brand named in a recorded model answer. Not a brand the model recognises when prompted directly, not a brand on a page the model cited, and not a brand a human decided belonged in the category. If it cannot be pointed at in a stored answer, it is not a mention.
That strictness is the whole basis for auditability. An index that cannot show you the sentence behind a number is asking you to take the number on faith.
Position, framing, and consistency
Three things separate a strong mention from a weak one. Position: named first carries more weight than named sixth, because buyers stop reading. Framing: recommended beats listed beats warned-against. Consistency: named on every run beats named once, because models are non-deterministic and a single naming can be sampling luck.
Scoring here weights position and consistency, and leaves framing to the recorded answer, for the reason set out under sentiment: collapsing framing to a number would add precision the evidence does not support.