AI Visibility Evaluation

Selection: being named when it counts

The only measure that touches revenue directly, and the one a business can fail after passing everything else.

What selection means

When someone asks a real buying question that this business should be an answer to, is it named? And when it is not, who is named instead?

Retrieval, comprehension and corroboration are all about whether the conditions for being recommended exist. Selection is whether the recommendation actually happens, and a business can satisfy all three and still not appear.

Why passing everything else is not enough

Selection is comparative. The other three are properties of the business; this one is a property of the field. A business can be perfectly readable, accurately described, and reasonably corroborated, and still lose to three competitors who are all of those things and better known. Nothing about the business changed. The comparison did.

There is no page two. A ranked list degrades gracefully: position eleven is worse than position three and still visible. A synthesised answer names two or three options. Everything else is absent, and absence is not a lower position, it is a different state.

Why the question wording decides the answer

Selection results are only meaningful against a question a real buyer would ask. Testing a business against its own brand name proves almost nothing: of course it is named. Testing against the category with no location returns national brands. Testing against the exact commercial question a customer asks is the only version that tells you anything.

This is also why selection numbers between tools are rarely comparable. Different questions produce different fields, and a favourable number is usually a favourable question.

Why it must be sampled, not asked once

Answers vary between runs. Asking once and reporting the result as a finding is reporting sampling noise. Selection has to be measured as a rate across many runs, reported with its sample size, and never presented as a single yes or no. A tool that gives you one answer has told you what happened one time.

Who is named instead

The more useful half of the finding. Knowing a business is absent is a problem statement. Knowing which three competitors are consistently named, and what they have that the business does not, is the beginning of an answer. Frequently the incumbent has nothing better except more independent corroboration, which points the work at exactly the slow block that most people want to skip.

Back to AI Visibility Evaluation