Glossary
What is answer volatility?
Answer volatility is how much an AI engine’s answer changes when you ask it the identical question again. It is measured as the overlap between the products named across two or more runs, and it is the reason a single-run visibility score cannot be trusted.
How is it measured?
Run the same prompt twice against the same engine, record every product named, and take the overlap: the share of products appearing in both runs. A category where both runs name the same five products scores 100%. A category where one run names two products and the next names five, sharing two, scores 40%.
How volatile are answers in practice?
Across ten WordPress plugin categories asked of Perplexity twice each on the same day, mean overlap was 74%. Translation and booking returned identical product sets. Backup returned 50% overlap and changed its top recommendation. Affiliate management returned 40%.
Volatility is a property of the category rather than of AI in general. That makes it a useful per-category difficulty measure: in a stable category, absence is a durable problem worth fixing. In a volatile one, a single absence may mean nothing at all.
Why does it matter commercially?
Because the noise is larger than the signal most vendors are trying to detect. At 74% mean overlap, a share-of-voice figure from one pass carries roughly a quarter noise — more than half in the worst category. If a three-month programme moves your visibility by ten points and your measurement error is twenty-five, the measurement cannot tell you whether the programme worked.
There is a second-order effect worth knowing: being named is more stable than being named first. Two categories returned perfect overlap on which products appeared and still changed which one led. A vendor tracking only mentions would call those categories solved.
What is the fix?
Multiple runs, spaced in time, with the variance reported alongside the average. Any AI visibility number quoted without a run count is an anecdote with a decimal point.
Part of the GEO glossary. The measurements referenced here come from the Plugin Visibility Index, which publishes its raw counts and its limits.