AI search visibility: an average hides where you are missing

AI search visibility is about how often your company is mentioned when someone asks ChatGPT, Gemini, Claude or Google's AI Overviews about what you sell. Many measurement tools sum this up in a single number. It might say, for example, that you are mentioned in 30% of answers, and the number is then followed month by month.

A number like that is an average. It mixes buyers with different needs and questions with different intent. A company can be highly visible when small customers ask and missing entirely when its largest customers compare suppliers.

My recommendation is that you measure visibility by audience and by intent, using questions your buyers actually ask. Then you can see where the gap is and do something about that gap in particular.

An example of what the average hides

An example: a made-up company sells coffee machines to offices. The company asks five AI services 60 questions and gets 300 answers. It is mentioned in 90 of the 300 answers, so the average is 30%. Here is how the same answers look when split by audience.

Audience	Answers that mention the company	Share
Small offices	39 of 75	52%
Office managers at mid-sized companies	27 of 75	36%
Property managers	21 of 75	28%
Procurement at larger companies	3 of 75	4%
All audiences	90 of 300	30%

The 30% average looks steady. The breakdown shows that the company is almost absent when procurement staff at larger companies ask, and that may be where the biggest deals are.

The same applies to the intent behind a question. The company may be mentioned when someone wants to know which suppliers exist and still be missing when someone asks for a single recommendation.

Answers differ from one time to the next

AI services do not answer the same way every time. In a study by SparkToro and Gumshoe, 600 volunteers asked ChatGPT, Claude and Google's AI tools twelve questions and collected 2 961 answers. For ChatGPT and Google's AI, the chance that two answers to the same question contained the same list of brands was less than 1%.

The researchers concluded that the share of answers in which a brand is mentioned is a reasonable metric when there are many questions and each is asked several times. A position in a single answer, on the other hand, says little.

Answers can also be shaped by what the service knows about the person asking. ChatGPT, for example, can use saved memories and past conversations to personalise its answers. Two buyers can therefore get different answers to the same question.

That is one reason each audience needs many questions. My assessment is that three questions per audience give a share that is too uncertain to compare over time.

How we build the questions

A measurement is never better than its questions. This is how we do it.

1. Choose four to five audiences. Start from your business areas and your search data. An audience is a type of buyer with its own need. It might be the procurement lead at a larger company or the owner of a small office.
2. Follow the buying journey. The questions move from finding out which suppliers exist to requirements, comparison and choice. A comparison question asks for the options to be weighed, and a recommendation question asks for a single choice.
3. Ask each audience the same number of questions. Around 15 questions per audience, spread across the stages of the buying journey, make the audiences comparable.
4. Anchor every question in the category. The question should include what you sell. The question "Which supplier offers same-day service?" could be about anything. The question "Which supplier of office coffee machines offers same-day service in Gothenburg?" measures the right market.
5. Let buyers supply the words. The search terms in Google Ads and Search Console are good sources, and so are the questions your sales team and customer service receive. Search Console is Google's tool that shows which searches your website appears for.
6. Leave your own name out. The questions should sound like a buyer looking for a supplier.
7. Ask the questions in several AI services and several times. ChatGPT, Gemini and Claude answer differently, and each one answers differently from one time to the next.

Read the answers behind the numbers

A mention can mean different things. The company might be the first choice in a recommendation or a name at the bottom of a long list. It might also be mentioned as an example of what the buyer should avoid.

So read the answers counted as mentions before the numbers go any further. Also look at which companies are mentioned instead of you in each audience and which websites the AI services refer to.

Keep AI search visibility separate from your position in Google's regular search results. They are two different metrics and each should have its own place in the report. The same principle applies to a dashboard, where an average or a bar without a denominator easily gives the wrong picture. I write about that in the post on a dashboard management actually reads.

How many visits actually come from AI services shows up in GA4. The guide to tracking AI traffic in Google Analytics 4 shows how to group them in a channel of their own.

What to do with a gap

A gap in one audience shows where the work should start. My assessment is that it is usually because the website does not answer that audience's questions, or because the company is missing from the sources the AI services read.

1. Read the answers for the audience and note which companies and websites are mentioned.
2. Compare them with the page on your website that ought to answer and see whether the answer is there.
3. Write the answer into the page with accurate facts about your offer.
4. Ask the same questions again after a quarter and compare the share for that audience.

Your next step

We measure AI search visibility by audience and stage of the buying journey, with questions built on your buyers' own words. We ask the questions in several AI services and read the answers before anything is counted. You see where you are missing and who is mentioned instead. For the largest gaps we write copy for the pages that should answer, and after about a quarter we measure again with the same questions. The measurement is most useful where buyers compare several suppliers before choosing, as in B2B.

Do you know which of your audiences cannot find you when they ask ChatGPT?