Start with a defined scope

A visibility percentage needs a denominator. Before counting anything, decide which questions belong in the study, which products you will check and what dates the study covers. Keep the exact wording of each prompt.

For a CRM team, separate broad discovery questions from migration, pricing and integration questions. A set made entirely of prompts containing your brand name tells you little about unprompted discovery.

Count distinct signals

Use separate columns for a brand mention, an explicit source citation and a visit to your website. A mention can appear without a link; a citation does not establish that anyone clicked.

Ahrefs also distinguishes mentions from citations in its own measurement definitions. When comparing reports, read each provider’s counting rules rather than assuming identically named metrics are equivalent. Source: Ahrefs: AI visibility metrics.

Keep the evidence

For each observation, retain the prompt, product identity, answer date, answer text and available citation metadata. If you save a source page, record its capture date separately. The page might have changed after the answer was generated.

• Record unavailable answers separately from answers with no brand mention.
• Count a brand once per answer for your baseline, even when its name repeats.
• Compare repeated observations with the same wording and collection context.
• Label every percentage with the number of observations behind it.

Use the result carefully

Suppose a brand appears in six of ten collected answers. That is 60% of this sample, under your counting rule. It is not market share, search demand or proof that 60% of people see the brand.

PromptFace lets you inspect the latest available public answer and its sources. For a historical study, agree the measurement period and available records before ordering a report. A change between two answers is an observation; investigate it before attributing it to an SEO edit.

• Turn answer data into a content brief