AI mention rate is a probability, and probabilities have confidence intervals. A single ChatGPT run that does not name your brand is not evidence; a sample of fifty runs where 32 named you and 18 did not is. AEO reports that quote a single number without a confidence band are either statistically naive or actively misleading — and that includes most public dashboards in the category as of 2026.
Why one scan is not a measurement
LLM answers are stochastic. The same question, asked twice, can produce different brand lists. Public benchmarks of ChatGPT and Claude on standardized AEO buyer questions show within-question mention-rate variance ranging from a few percentage points to over twenty, depending on the category. A single scan is therefore one draw from a distribution. Reporting it as the mention rate is like reporting one die roll as the average roll.
How many samples you actually need
For a binary outcome (mentioned or not), the standard formula gives a 95 percent confidence interval roughly proportional to 1 over the square root of the sample size. At 25 samples per question per model, a true 50-percent mention rate has a confidence band of plus or minus 20 points. At 50 samples, it tightens to plus or minus 14. At 100 samples, plus or minus 10. To detect a real 10-point change between scans, 50 samples is the practical minimum.
Why competitor variance also matters
Reporting your mention rate without your competitors' is a one-sided view. You might be at 40 percent and steady, while the leader dropped from 70 to 50 — a relative gain you would miss. Always report mention rate with the same confidence interval applied to the top three competitors on the same question and the same scan. The relative position is the load-bearing number; the absolute number is the supporting detail.
Reading week-over-week movement correctly
A 5-point week-over-week move at 50 samples is well inside the noise band — it tells you nothing. A 12-point move is plausibly real and worth investigating. A 20-point move is almost certainly real and should kick off a citation-source audit immediately. The discipline is to draw the noise band on every chart and treat moves inside it as not signal, regardless of how dramatic they look.
What AskRanker reports by default
Every AskRanker mention-rate number ships with a 95 percent confidence interval based on the sample size used. Charts show the band visually; numerical reports show the plus/minus alongside the point estimate. Significance markers flag movements that exceed twice the standard error. This is the minimum viable level of statistical rigor for a metric that drives expensive content investment, and most of the AEO category has not gotten there yet.