AI audit sample size planner
Find out how many matched AI answers to compare when you want to estimate how often a brand mention or official link changes.
Your collection target
139
matched pairs
That’s 278 saved answers if each answer is used in only one pair.
Budget for more runs if you expect missing answers or will repeat the same prompts.
How to use the target
Pick one outcome to track, such as a brand mention appearing or disappearing. A matched pair compares the same prompt, engine, settings and market across two collection rounds. Aim for a mix of distinct prompts. Repeated comparisons of the same prompts or adjacent days can be related and give you less information.
The VisibilityOS study observed 82 brand-presence changes across 817 matched pairs, about 10%. The calculator uses 10% as its starting assumption. Rates will vary across audits. See the study and method.
How the estimate works
For an expected rate p and margin e, both as decimals, round z² × p(1 − p) ÷ e² up to a whole pair. The confidence level sets z. This approximation assumes independent pairs and can be unreliable when few changed or unchanged pairs are expected. Use it as a planning estimate. NIST sample size guidance.