The three measured rates
These definitions describe the Category Audit's three reported measures. Named and cited as a source use the valid captured answers. Share of voice uses the brand mentions recorded in those answers. Each has its own count and denominator. We never add or weight the measures into a composite score.
Named
An answer counts once when it names the company. The denominator is the valid answer base for the total, engine or question group being reported. A full Category Audit schedules 150 observations: ten questions on ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, with three runs of each question. Failed or empty observations are excluded from the answer base and reported separately.
Cited as a source
An answer counts once when at least one listed source is on the company's own website domain or a confirmed alias, including its subdomains. It uses the same valid answer base as named. The cited source list is the evidence; we do not infer a citation from an uncited mention in the answer text.
Share of voice
The numerator is the answers naming the company. The denominator is the mentions of the company and the recorded rivals across the same valid answers, with each brand counted once per answer. This covers the brands recorded for the order, not every brand in the market. It is a share of brand mentions, not a percentage of answers.
Record accuracy, when classified
The report identifies which company or recorded rival each classified statement describes, and counts statements classed correct, wrong or stale against the dated public evidence. An unverified or unclassified statement is not automatically wrong. A classification remains provisional until its team review is recorded. When no classification exists, the report does not imply a zero error count.
Attribute profile, when classified
Recorded descriptions of the company and its rivals carry the number of supporting answers and a receipt. Unreviewed rows are labelled: “From the captured answers; provisional until reviewed.” The profile does not imply that every rival or every possible attribute was assessed.
Scope and related rules
These descriptive measures are separate from the Fix Sprint payment tests in the current terms. Those tests use named answers, cited as a source and record accuracy; share of voice is not a payment test. The wider 35 question program method and historical measurements retain their own scope.
The Kalvenor findings sample uses fictional SAMPLE DATA to show the reporting format. Its figures do not describe a real customer or a future result.
The separate bot access resource explains the limits of server requests and HTML checks. These rates measure captured answers; they do not establish why a source was chosen or what a buyer did next.
Every figure comes from the published method, v1.1.