Should AI visibility tests name the business in the prompt?

While building AI‑Tracker, I encountered a methodological problem.

Many AI visibility tests ask something like: “What do you know about [business name]?” That measures recognition, but it doesn’t establish whether the AI would independently recommend the business when a customer asks who to choose.

Our approach is to ask realistic customer questions without naming the target business, then measure whether it appears naturally in the answers.

However, blind testing also introduces complications. Different AI providers may interpret the same question differently, and some answers provide general advice without recommending any supplier.

How would you define a fair AI recommendation test?

Should recognition and blind recommendation be reported separately, or is there a better methodology?