Why is n=1 not enough?

At one sample you have no interval at all. Repeating the same prompt commonly returns a different citation set, so a single reading tells you what happened once, not what happens typically.

What sample size do you default to?

Ten per prompt per engine, which is a deliberate cost compromise. For change detection on important head terms, twenty-five is a better default.

← All free tools

AI answer sample size calculator

AI answers are non-deterministic, so citation share is a proportion estimated from n samples. To detect a 10-point change at 95% confidence around a 30% share you need roughly 81 samples per prompt; at n=10 your interval is about ±25 points, which is wide enough that most reported movements are noise.

Expected citation share: 30%

Change you need to detect: 10 points

Samples required per prompt

At 95% confidence, to resolve a 10-point change around a 30% share.

Interval at common sample sizes

Samples	95% interval
n=10	±2.8 points
n=25	±1.8 points
n=50	±1.3 points
n=100	±0.9 points

Runs entirely in your browser

Nothing you type here is sent to us or stored anywhere. Copy the output and go.

Why is n=1 not enough?: At one sample you have no interval at all. Repeating the same prompt commonly returns a different citation set, so a single reading tells you what happened once, not what happens typically.
What sample size do you default to?: Ten per prompt per engine, which is a deliberate cost compromise. For change detection on important head terms, twenty-five is a better default.

Other tools

AI robots.txt generator llms.txt generator Prompt set builder