How to measure AI visibility across ChatGPT and AI search.
AI visibility is measured by observing whether, where and how a brand appears across a controlled set of questions and answer systems.
A single screenshot is an observation. A repeated method creates a baseline.
Three measures that are easy to confuse, and shouldn't be.
Mention rate
How often the brand is named across the defined question set.
Citation rate
How often an answer points to the brand's site as a source. A citation is not automatically a recommendation.
Share of answers
How often the brand appears relative to the competitors observed across the same test.
Technical readiness may be reviewed alongside these measures, but it is an input, not a visibility outcome.
Five steps, run the same way every time.
Platforms, prompts, markets and competitors.
Full answers, cited sources, date and context.
Mentions, roles, citations and description accuracy.
Separate observed evidence from interpretation.
Use the same core test to observe change.
Prompt sets should reflect real customer questions. Repetition captures variance instead of hiding it. Dating and preserving the full answers makes later comparisons reviewable.
Observation and interpretation, deliberately separated.
Measurement tells you what the systems returned. Not why.
Relevant evidence is not being found.
The brand, service or location is misinterpreted.
Important claims lack support.
Another brand has a clearer retrievable trail.
How the measurement approach described here separates documented platform behavior from Bnarrativ's own observation, analysis and limits.
Google documents that its AI features may issue multiple related queries — described as query fan-out — and draw on core Search systems (Google Search Central: AI features and your website).
Bnarrativ records mentions, citations, recommendations and competitor appearances for a fixed prompt set, then repeats the same prompts over time. See the public Cited protocol.
A single answer is an observation; repetition across prompts, systems, markets and dates is what turns observations into a defensible baseline.
AI answers vary by platform, prompt, model, market and date; a baseline describes patterns, not deterministic rules, and cannot guarantee future placement.
What is the difference between a mention, a citation and a share of answers?
A mention is how often the brand is named across your question set; a citation is how often an answer points to your site as a source; share of answers is how often you appear relative to the competitors observed across the same test. They are easy to confuse and should be measured separately.
Why does the denominator matter?
A visibility figure means little without knowing how many prompts, on which platforms, over what dates it is drawn from. Disclosing the denominator is what makes the number reviewable rather than a headline.
Should I name my brand in the test prompts?
No. Prompts that name the brand tend to surface it by default, which inflates the result. Prompt sets should reflect the real questions customers ask, so the observation reflects how people actually search.
Establish a baseline before choosing the remedy.
Cited turns controlled prompt testing, citations and competitor patterns into evidence an analyst can review.
Explore the Cited AI Visibility Audit →