Technical explanation
Generative models sample from a probability distribution, so brand mentions and citations can appear in one run and vanish in the next. Reporting a single run as a fact is the most common measurement error in this field.
Example
Five runs of the same prompt might mention your brand three times — that is 60% visibility with real uncertainty, not a fixed rank.
How it differs
Classic rank tracking assumes a stable result. AI answer tracking must assume the opposite.
Related terms
Where Varazm measures this:
AI Visibility MonitoringMethodology