Every AI visibility tool infers from sampled responses. Most do not say so.
A fixed panel of buyer-intent prompts per category, run across engines on a schedule, each prompt sampled several times because identical questions produce different answers. One response is an anecdote, not a measurement.
Panel questions are immutable, so when a number moves you know the models changed and not the question.
When you ship a change, we measure your movement against competitors in the same panel over matched windows, and report the effect with its uncertainty. Anyone claiming to prove causation here is guessing.
Nobody outside a model provider can watch how it selects sources. We infer from what the answers cite, and we tell you that.
AI referrals get undercounted: zero-click answers leave no trace and referrers get stripped. We report a range with a stated floor rather than a confident total.
This disclosure is a commitment, not copy. It appears on the marketing surface because the limitation is the trust argument.
