Sign in

AI visibility

Updated: 3 min read SEOFuxx editorial team

AI visibility describes how often and in what form a brand, product or website appears in the answers of AI systems. It is the counterpart to classic visibility in search results, but it is not measured through positions but through mentions and source citations in answers.

Metrics

It is measured by regularly putting a fixed list of typical questions (prompts) to AI systems and evaluating every answer. Common metrics:

MetricMeaning
Mention rateShare of answers in which your brand is named
Citation rateShare of answers that link your website as a source
Share of voice, share of modelYour share of all brand mentions when you and your competitors together make up 100 percent (see share of model)
List positionWhere your brand stands when the answer contains an ordered list of recommendations
SentimentWhether the portrayal is positive, neutral or negative

Every metric comes with the number of answers it was calculated from, for example "7 of 12 answers". Without that information a value cannot be put into context.

Why only samples are possible

AI answers are not deterministic. The same question can be answered differently on every query, depending on the model, the web search that is retrieved and on language and location. Every measurement is therefore a sample. An example shows how large the uncertainty is: if your brand is named in 7 of 12 answers, the actual rate lies, with the usual statistical confidence, somewhere between about 32 and 81 percent.

The values only become reliable through repeated measurements. Ask each prompt several times and watch the development over several weeks instead of comparing single runs.

How a clean measurement is set up

  • Fixed prompt set: Questions the way your audience asks them, mixed from information, purchase intent, comparison, local reference and brand questions.
  • Fixed parameters: Platform, model, language and country stay the same within a run. If a parameter changes, the runs are no longer comparable.
  • Repetitions per prompt: Several queries per question smooth out the fluctuation.
  • Brand recognition: Name, domain and spellings have to be recognized reliably. If the brand is named like a common word, an additional criterion is needed, such as the domain being named.
  • Competitors for comparison: Only the gap to competitors makes a rate meaningful.
  • Store raw answers: Only that way can every metric be traced back to the original answer.

Limits of what it tells you

  • The prompts are a selection. How often people actually ask a question is not known.
  • Model and index updates by the providers can change values overnight without you having done anything.
  • Personalization and location influence answers. A measurement run only shows one constellation.
  • A mention is not a visit. How many visitors come from AI systems you see in web analytics (see AI referral traffic).
  • A single overall value (such as "score 72") hides where the number comes from. The individual values with their sample size are more meaningful.

Improving visibility

Measures that raise the chance of mentions and citations are described in the article on GEO. The measurement shows which prompts never reach your brand (the "gaps"), which sources the systems use instead and how your values develop after changes.

Common mistakes

  • Asking once and judging: A single answer in ChatGPT says little about actual visibility.
  • Over-interpreting small samples: With few answers a value jumps strongly.
  • Changing parameters in the middle of a comparison: Anyone who switches country, language or model is no longer comparing the same thing.
  • Looking only at your own brand: Without competitors the yardstick is missing.
  • Counting false mentions: Common words and similar-sounding brands distort the rate.

Sources

Is this content helpful?

·