A brand can appear in an answer without being presented as a suitable choice. It may be background context, an alternative to avoid or one name in a long list. A mention count alone cannot tell these situations apart.
My approach starts with the buying situation: who is asking, what do they need and what constraints actually matter? Only then does it make sense to examine which options the answer recommends and why.
A useful measurement system separates presence, recommendation and accuracy. It also records uncertainty. That produces a more actionable conversation than a single visibility number.
The practical next step: choose a small set of commercially meaningful questions, define what counts as a recommendation and review the answers before turning them into a score.