Measuring AI brand visibility: questions and answers

How is an AI visibility percentage calculated?

Contxt editorial

For a mention-based measurement, divide the number of tested answers that name the brand by the number of eligible, successfully collected answers, then multiply by 100. Always read the denominator and collection scope alongside the percentage.

If 58 of 200 tested answers mention a brand, the measured visibility is 29%. This is an illustrative calculation, not a customer result or a claim about the brand’s share of all AI conversations. If the test used 50 prompts across four platforms, each prompt-platform answer is a separate observation.

Platform and buying-stage breakdowns can reveal what the overall score hides. A 29% aggregate could include strong coverage on one platform and little coverage on another. Similarly, a brand might appear in comparison answers but be absent from early research questions.

A failed request, an unavailable answer and an answer that genuinely does not mention the brand must not be treated as the same thing. Review collection completeness before interpreting a change. Rankings, sentiment and cited source counts are separate measures with their own definitions.

For repeat comparisons, check the prompts, platforms, dates and collection conditions. A changed prompt set can produce a changed percentage without demonstrating an improvement in the same buyer questions. See comparing visibility over time and Contxt’s reporting features.

Why do ChatGPT, Google AI Overview, Claude and Perplexity give different results?

Contxt editorial

Platforms can use different models, retrieval systems, source selections and answer formats. Results can also change with the question, collection date and context. Google does not show an AI Overview for every search.

Read each platform separately before relying on an aggregate. A result on one assistant is not proof of coverage on another. Contxt’s public visibility workflow covers ChatGPT, Google AI Overview, Claude and Perplexity; check the recorded answer and collection status for each.

How should I compare AI visibility over time?

Contxt editorial

Compare repeat scans using the same prompts and platforms wherever possible. Record when content changes were made and examine which individual answers changed, rather than judging success from the overall percentage alone.

If you change the prompt set, keep both readings but label the difference in scope. Same-day runs can differ too, so a single increase does not prove that a content change caused it. Look for repeated evidence across the relevant buyer questions and use website analytics to assess traffic or conversions separately.

Does zero visibility mean the scan failed?

Contxt editorial

Not necessarily. A valid zero means answers were successfully collected and none mentioned the brand. Missing data means the collection did not provide a usable reading. The two need different explanations.

Check the tested-answer count and platform status. Do not interpret a failed or empty collection as evidence that the brand disappeared. Conversely, do not hide a valid zero: it identifies questions where the recorded answers currently do not name you.