Why sample size is the number that matters most

Measuring visibility is polling, not ranking

Because AI answers are probabilistic, measuring visibility is closer to polling than to checking a rank. Ask an engine a question once and you have an anecdote. Ask it many times, across models and conditions, and you have an estimate — with a margin of error that shrinks as you ask more. The sample size, n, is what tells you which of those two you are holding.

Why we write it as θ

We write visibility as θ (theta) — statisticians' notation for a probability you are estimating, not a fact you have looked up. Writing it that way is a standing reminder: the honest object is 'the estimated chance an answer mentions you', and it should never be reported without the n that supports it.

The failure mode: small-n confidence

A brand mentioned in four of five answers looks like an 80% star — until you notice the next five samples land at 40%, and the 'trend' was noise all along. Small samples swing wildly; the fix is not a cleverer chart but more data, and the discipline to widen the error bars honestly when the data is thin.

Sample size guards against cherry-picking

With enough small, hand-picked prompt sets, almost any story can be told. A disclosed, stable, reasonably large sample makes that far harder — the denominator is the honest constraint that keeps a measurement from becoming a marketing exercise. When you read any AI-visibility claim, the first question is not 'how high?' but 'out of how many?'