Calculate the observed rate
Enter the counts from a frozen test log. The calculator treats each eligible AI answer observation as one opportunity and keeps mention rate separate from visible own-domain citation rate.
Add counts to generate a sentence you can paste beside the prompt set, product, dates, and exclusions.
What the calculator means by citation rate
If 20 eligible responses were recorded and 5 visibly cited the measured domain, the observed own-domain citation rate is 25% for that test.
Mention rate uses the same denominator, but counts eligible answers that named the brand whether or not they visibly cited the domain. Keeping those two rates separate prevents a single visibility number from hiding what actually happened.
Why the interval is included
The calculator uses a 95% Wilson score interval for a binomial proportion. This is a statistical interval for a defined set of observed opportunities. It is included because a rate from 5 of 20 runs should not be reported as if it has the same precision as a rate from 50 of 200 runs.
Sources: NIST Engineering Statistics Handbook, confidence intervals for proportions and Brown, Cai, and DasGupta, Interval Estimation for a Binomial Proportion.
How to report the result
- Publish the exact prompt set, product, market, mode, dates, account state, and run count.
- Report excluded runs separately from eligible runs.
- State the numerator and denominator beside the rate.
- Keep mention rate, own-domain citation rate, and accuracy review separate.
- Keep the raw observation log available for review.
For collecting the raw observations before calculating a rate, use the AI citation test workbook. For choosing a run count, read how many runs an AI citation test needs.
What this calculator does not prove
The output is not a ranking, forecast, causal proof, or permanent visibility claim. AI answer behavior can vary by product, time, market, account state, visible mode, interface, and retrieval behavior. Repeated AI answers may also violate simple independence assumptions, so the interval should be treated as measurement context for the observed test rather than a guarantee about future answers.
Audit the source page before retesting it.
A calculator can summarize observations. It cannot fix the page behind them.
Audit a page free