Concept · How AI answers work
Answer variance
Definition
The tendency of an answer engine to give different answers to the same question.
What it means in practice
Sampling, retrieval, and personalization all move between runs, so a single answer is one draw from a distribution rather than a reading of a fixed rank. A brand present in one answer and absent from the next has not necessarily changed. The practical consequence is that any AI visibility number built on one observation per question is not measuring the brand, and that comparisons need repeated runs over a fixed prompt set.
How refd treats it
This is category vocabulary, not a refd metric. refd measures what AI answers actually said across repeated runs, and publishes the raw answer behind every number. Where a term above carries a claim about what causes visibility, refd treats that claim as a hypothesis to test rather than a setting to configure.
This definition is editorial. It describes how the term is used across the field rather than something the software computes, so unlike the metric pages it is not generated from the product.
See your evidence