Skip to main content

How do you measure whether AI assistants are mentioning your business?

AI Search Published November 2, 2026
Short Answer

There is no rank tracker for generated answers, so you measure with a panel. Build a fixed list of questions a real customer would ask, run them on a schedule across the assistants that matter, in clean sessions, and record whether you were mentioned, whether you were cited with a link, and who appeared instead. Track the share of the panel where you appear. It is sampling, not measurement, and it has to be treated that way.

Design the panel like a survey, not a keyword list

Pick questions across the buying journey: symptom questions, repair-versus-replace questions, cost-context questions, hiring questions, and direct brand questions. Forty to a hundred is enough for a single market.

Then freeze it. The moment you start swapping questions in and out, your trend line becomes meaningless, because you changed the instrument and the subject at the same time. Add new questions as a separate cohort with its own baseline.

Include questions you expect to lose. A panel built only from questions you already win produces a flattering number that never moves and teaches nothing. The value is in watching which categories are winnable and which are permanently owned by a national source.

Control what you can, and write down what you cannot

Run logged out, in a fresh session, with location stated explicitly in the prompt or in the client. Note the assistant and, where visible, the model version. Run each question more than once, because outputs vary between identical runs.

Record the raw response text, not just a yes or no. Six months later the interesting question will be how you were described, not whether your name appeared.

Be honest in reporting about what this is. It is a sample of a nondeterministic system, run from one vantage point, at one moment. It supports statements about direction. It does not support precise share claims, and any vendor presenting one without describing the sampling method is describing something else.

What to record for each run

  • Mentioned by name - appeared in the answer text at all.
  • Cited with a link - the stronger signal, and the one that can produce a visit.
  • Source URL used - frequently a directory or review site rather than your own domain, which is itself the finding.
  • Competitors named - who is occupying the answer you want.
  • Description accuracy - whether the services, area and positioning attributed to you are correct.

Read the trend, never the reading

Variance between runs is large enough that a single day's result means very little. Look at multi-week direction and at aggregate appearance rate across the panel, and report the number of observations alongside the rate so nobody mistakes a sample of twelve for a fact.

Because this is manual by nature, it is worth automating the collection and putting the history somewhere durable - the same reason we build reporting dashboards rather than emailing spreadsheets. The panel result belongs next to organic sessions and booked jobs in marketing intelligence, not in a separate deck.

Pair every panel review with one qualitative read. Take five answers where a competitor appeared and read what the cited source actually said. That usually explains the number faster than the number does.

Topics: measurement · visibility · sampling · reporting

Have a version of this question about your own business?

The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.

Related Answers

People who read this also asked

Browse the Answer Hub →

AI is easy to access. Making it useful is hard.

Bluefrog makes AI useful by integrating it with the way your business actually works — your software, your calls, your customers, your marketing and your revenue.

Technology development since 1997 · AI integration platforms since 2001