How to audit AI visibility for a client: a working guide

Decide what you are auditing first

Most AI-visibility audits go wrong at the first step, by asking the wrong kind of question. Settle three things before you ask anything.

• The category, in the customer’s words. Not “IT solutions” — “managed IT and cybersecurity services”. Assistants answer the question asked, and nobody asks for solutions.
• The place, if the business is local. “Best plumbers” returns national chains and is useless. “Best plumbers in Denver, CO” is the question a customer actually types.
• The brand terms you will count as a mention. The legal name, the trading name, the domain. Get this wrong and you will record a miss for an answer that named the client.

The five questions worth asking

Five phrasings of one buying intent gives you a rate rather than an anecdote, without the cost of thirty. Vary the phrasing, hold the intent constant.

• Who are the best [category] providers in [place]?
• What companies offer [category] in [place]?
• Recommend a reputable [category] company near [place].
• Which [category] firms have the best reputation in [place]?
• I need help with [category] for a small business in [place]. Who should I call?

Do not ask “what do you know about [client]?” as your visibility question. It tests recall, not recommendation, and it almost always returns something — which makes a business look visible when no customer would ever have found it.

Reading the answers

Record four things per assistant. Each one implies different work, which is the point of separating them.

Known but never recommended is the common result

Most established businesses are recognised by assistants and recommended by none of them. That is not a branding problem and it is not fixed by publishing more pages about yourself. It is usually a sources problem: the assistant is drawing its recommendations from directories, roundups and review sites the business does not appear in.

The part clients actually react to

Write down every company named in place of your client. That list is the most useful output of the audit, and it is the one that ends the meeting with a decision.

Then go further than the list: grade those competitors’ websites on the same rubric you graded your client’s. It converts an abstract complaint — “AI does not mention us” — into a comparison with numbers, and it very often shows the recommended companies scoring worse on the fundamentals. That is a much easier conversation than telling a business it is invisible.

“The four companies AI recommends instead of you average 62. You score 81.” That sentence sells work. “You have an AEO problem” does not.

What to hand over

A useful AI-visibility audit fits on two pages and contains:

• The exact questions asked, verbatim — not a summary of them.
• Per assistant, named or not, for each question. Never a blended score.
• The four measures: awareness, accuracy, reputation, recommendation.
• Every competitor named instead, with their own scores beside the client’s.
• Three actions, ranked, each tied to a specific finding.

Anything a client cannot check themselves will eventually be checked and found wrong. Showing the questions verbatim is what makes the rest of the document credible.

How often to repeat it

Monthly. Assistants that answer from live search move as sources move, so a monthly re-check produces a trend line — and a trend line is what justifies a retainer rather than a one-off project.

You can run the measurement described here free on any domain, with no signup, using our AI visibility checker, or automate the whole audit including the competitor scoring with ActiScore.