What Does AI Visibility Actually Measure?

GrayRhino AI separates completion coverage, brand discovery, brand descriptions, owned citations, and competitor judgments. Each metric preserves its denominator, unknown count, and excluded count.

Completion coverage uses the expected question-by-source task matrix. Brand and owned-citation ratios use only completed, relevant, and decidable tasks; unknown evidence is never silently converted to a miss. The report also retains the completed population, unknown count, and excluded count so the reader can see why a task was not in a particular denominator.

Source structure counts valid citation links rather than answer tasks. Competitor lead counts applicable comparison questions after a question-level judgment, not the number of provider answers. The ratios may appear together in a report, but they do not share a unit.

Synthetic teaching example: four base questions and two sources create eight expected tasks. Five complete, two fail, and one is missing, so completion coverage is 5/8. Among the five completed answers, suppose one is off topic and one has an undecidable brand identity. If the remaining three are relevant and decidable, a single confirmed brand mention yields 1/3 for brand discovery, not 1/5 or 1/8.

If the three relevant answers contain four valid citation links, source structure starts with four links as its population. Calling this 'four of eight answers had sources' would change the unit. These values are invented solely to show the calculation.

Each response stores relevance, brand mention, owned citation, source class, and competitor judgment independently. Deterministic rules then calculate numerators, denominators, unknowns, and exclusions. A brand description stays attached to its answer sentence and task rather than being merged into an invented cross-source consensus.

A mention only proves that the brand appeared in one answer. Description accuracy, source ownership, and relevance require separate evidence.