How to measure AI brand visibility.

1. Define a specific market

Use a product category and target audience that describe actual buying intent. “CRM software for small UK businesses” is more useful than “business software.” Avoid including your company name in discovery questions: doing so biases the test.

2. Keep the conditions consistent

Record the prompt, model identifier, date, and response. Use the same questions when repeating an audit. API model responses and consumer apps are different testing surfaces and should not be combined as if they were the same.

3. Separate mentions from recommendations

Our score is the share of completed answers containing your company name or domain. It does not measure recommendation quality, sentiment, market share, or traffic. Read the answers to check ambiguous names, negative mentions, and unsupported claims.

Example: 2 mentions in 6 completed answers = 33% after rounding. Failed responses are excluded, never counted as absence.

4. Review the evidence

Look at competitor mentions and any citations returned by the model. Verify claims at the source. A missing citation does not prove that a model did not use web information, and a citation does not guarantee that the answer accurately represents its source.

5. Retest without overclaiming

Six questions per model form a small exploratory sample. Responses may vary. Broaden your question set and repeat runs before drawing conclusions. Use the findings to improve clear product descriptions, buyer-facing documentation, and verifiable customer evidence. No specific ranking increase is guaranteed.