Resources Free tool
Run each prompt 93 times to report a mention rate within ±10 points.
At 95% confidence and an expected rate near 50%. A sampled AI answer is one draw, not a reading; the number of runs sets how wide the interval around your rate will be. Computed with the exact Wilson half-width, the same function the product uses.
Precision you need
Confidence
50% is the worst case and needs the most runs. If you do not know, leave it there.
Defaults: the demo workspace’s prompt set and cadence.
How the interval narrows at 50%
The same rate at four sample sizes. Precision grows with the square root of n: four times the runs halves the interval.
For your prompt set
- Runs per prompt
- 93
- Weeks at your cadence
- 19 (5 runs a week)
- Runs per engine
- 1 674
- Across 7 engines
- 11 718
Pool weeks until each prompt reaches 93 runs; the product does this by default.
Runs needed by precision, 95% confidence
Half-width of the interval by sample size
Wilson score interval. At 50% this is the worst case.
Notes
The variance of a proportion is p(1−p), largest at p = 0.5. A brand named in roughly half of answers needs more runs than one named in a tenth, for the same precision. When you do not know the rate, plan for 50% and the interval can only be narrower than planned.
Pooling raises n, but only over runs that measure the same thing. Prompts with the same intent on the same engine pool well; runs across engines do not, because each engine cites and recognises brands in its own way. The product reports per engine and pools weeks and prompts within an engine.
With 5 runs per prompt a week, one week’s interval at 50% is about ±33 points. That is enough to track direction over months and far too wide to call a weekly change. A change is reported only when a two-proportion test says the difference is unlikely to be noise.
The method behind this calculator is written out in Measuring AI visibility honestly .
See these intervals on a real prompt set.
The demo workspace samples a fictional brand across seven engines, five runs per prompt per week, and shows every rate with its interval and n.