Wilson confidence interval
A Wilson confidence interval is a range around a measured proportion that reflects how much the sample size limits certainty. It is the standard tool for small-sample rates, and it is why an honest AI visibility figure is reported as a band rather than a single number.
Why not the simple formula
The textbook normal-approximation interval behaves badly exactly where AI visibility measurement lives: small samples, and proportions near zero or one. It can produce a lower bound below zero, or a zero-width interval when a brand appeared in none of the runs, which would assert perfect certainty from very little evidence.
The Wilson interval stays inside zero and one and keeps a sensible width at the extremes. A brand named in none of twelve runs does not get a confident zero. It gets zero with an upper bound that honestly reflects twelve runs being a small number.
What it changes in practice
It changes what counts as news. If a rate moves from 31 percent to 38 percent and the two intervals overlap, the honest report is that nothing has been demonstrated yet. A lift is claimed only when the ranges separate, and until they do the right word is trending, not won.
It also sets a floor on useful sampling. Very small runs produce intervals so wide that almost no change could ever be significant, which is good reason to be suspicious of any tool reporting a daily figure from a single answer per prompt.
Related terms
- Named rateNamed rate is the share of sampled AI answers that mention your brand at all, decided by exact word-boundary matching rather than by a model's judgment.
- Share of voice in AI answersShare of voice in AI answers is the proportion of answers in a category that mention your brand rather than a rival.
- Prompt trackingPrompt tracking is running a fixed set of buyer questions through AI engines repeatedly and recording what each answer says.
Want this measured on your own domain rather than defined? Run a free report and see the rates, with their sample sizes and ranges, for your own buyer questions.