Try Cited for free
7-day trial

How to measure AI visibility

By
K—lab
·
November 19, 2026
·
10
min read
How to measure AI visibility

How to measure AI visibility: ask ChatGPT, Copilot and Google AI Overview the same buyer questions several times, and count how many of the answers mention, recommend or cite your company. A number is only worth trusting once you can see the numerator, the denominator and the date, and once you have checked that the AI is talking about the right company.

It sounds simpler than it is. One question in ChatGPT gives you a snapshot, not a measurement. Below you will find six numbers to calculate, how many questions you need, and what we have learned from measuring AI answers every other day. We explain what AI visibility is in What is AI visibility? This article is about how to measure it.

What should you measure? Six numbers

Measure six numbers, and always write down the numerator and the denominator. A number without a denominator cannot be compared with anything.

MetricNumeratorDenominator
Mention rateAnswers that mention your brandAll answers to questions without a brand name
RecommendationsAnswers where the brand is put forward as a choiceAnswers to questions where the AI chooses or compares vendors
CitationsAnswers that link to your domainAll answers
Share of voiceMentions of your brandMentions of all brands in the same answers
ToneAnswers that describe you positively, neutrally or negativelyAnswers that mention you, not all answers
PositionYour brand's place in each list, added upNumber of answers with a ranked list

Mention rate is the main number. It tells you whether the AI knows you at all. Position is the average of the places, and only makes sense where the answer actually ranks.

Branded questions must be counted separately. When the question mentions you, the answer almost always does too. In our own measurement on November 18, 2026, almost every answer to questions containing the name K—lab mentioned us. On questions without the name, it was only about 2%. Mix the two and you get a number that says nothing about either.

Keep the platforms apart as well. Google AI Overview does not appear on every search, so it has its own denominator. You can read more about share of voice in Share of voice in AI.

How many questions and repetitions do you need?

You need 20 to 50 buyer questions without brand names, and you should ask each of them more than once. With few questions the denominator gets small, and small denominators give uncertain percentages (more on that below).

How to build the list:

  • Write the questions the way a customer writes them, not the way you describe the service yourself.
  • Cover the whole buying journey: research (“What should we look for when choosing a vendor of ...?”), comparison (“What is the difference between ...?”) and decision (“Who has experience with ... in Norway?”).
  • Leave out your brand name. Questions with the name go in a separate series.
  • Include questions where competitors are named, so you can calculate share of voice.

Our own setup is one example. K⁠—⁠lab puts a fixed set of buyer questions to three platforms, and the measurement runs every other day. Each question is asked once per platform per run, so the repetition comes from running the measurement again. If you measure by hand, you can ask each question three times in new chats on the same day.

Why do you need repetitions? Because the answers vary.

Why do AI answers vary from one measurement to the next?

AI answers are not fixed. The same question on the same platform can give a different answer, because the model chooses among many possible wordings, and because the tools run new web searches behind the scenes (see what query fan-out is).

We have measured how much. Between November 4 and 18, 2026, we compared each measurement with the previous one, for the same question on the same platform. That gave thousands of pairs. In under 10% of them, the set of brands mentioned changed, looking at the five brands we track. In over half of the pairs, neither answer mentioned any of the five. If you look only at the pairs where at least one answer mentioned a brand, the list changed in over a fifth.

That means two things. One measurement says little, and a change of a few percentage points between two measurements is often noise. And small numbers are uncertain: with a small denominator you cannot tell 2% from 4%.

How do you avoid counting the wrong company?

Check that the AI is talking about the right company before you count a mention. If you count every hit on the name, you overstate your visibility when someone else has the same name.

We know this from our own measurement. The name K-lab is also used by a gas metering test center at Kårstø. On November 4, 2026, Google AI Overview answered the question “What questions should we ask K—lab before we sign an agreement?” (asked in Norwegian). Its answer, translated: “Before you sign an agreement with K-Lab (Kårstø Metering and Technology Laboratory), you should ask specific questions about test conditions, accreditation, capacity and handling of raw data.” That is not a mention of us. It is a mention of another company.

From November 4 to 18, 2026, the name K-lab appeared in the text of about 15% of the answers. Only a handful of them mentioned the test center at Kårstø. Half of those were about the wrong company and were not counted as mentions of us. In the rest, the AI itself wrote that the name can be confused with the test center, and the question mentioned K—lab.

It is a small number of hits, but without the check the number would have looked too good. Here is how to do it yourself:

  • Read a sample of the answers that mention you, at least 20.
  • Check whether the description matches your industry and your service.
  • Remove the hits that refer to another company before you calculate the numerator and denominator.

What is the difference between “not measured” and 0%?

0% means you were measured and not mentioned. “Not measured” means you have no data, and then the number should be left blank, not shown as zero.

Example: on November 18, 2026, Google AI Overview answered almost all the questions. The questions without an AI Overview are not a “no.” They are not measured, and they do not belong in the denominator for Google. The same applies to an answer that failed for technical reasons, or a platform you have not measured yet.

If “not measured” turns into 0%, it looks as if visibility dropped, even though the measurement was simply missing. K—lab shows a blank where there is no data, and never 0%.

Should you measure by hand or with a tool?

Measuring by hand is fine to start with. With a tool you do not have to handle the repetitions, the history and the right-company check yourself. As of November 19, 2026, the options look like this, according to the companies' own websites.

OptionPlatformsPrice and trialWhat you should know
ManualAny you open yourselfFreeOne answer at a time. You keep track of the numerator, denominator and date yourself, and get no history.
K—lab Cited (we make it)ChatGPT, Microsoft Copilot, Google AI OverviewFrom NOK 5,900/month (see pricing). 7-day trial. Free checkShows numerator and denominator. You can open each answer and see sources and competitors.
Synlig.aiChatGPT, Google AI Mode, Gemini, Claude, Copilot, PerplexityStarter NOK 799/month (one model, 50 prompts). Pro NOK 2,899/month (two models, 100 prompts). Excl. VAT. 7-day trialVisibility score, website score and competitor ranking.
Sikte AIChatGPT, Perplexity, Gemini, Google AIFree visibility check. Monitoring price not statedDescribes itself as a Norwegian agency, not software. Monthly reporting.
GeoStarsChatGPT, Google AI Overviews, Gemini, Perplexity, Copilot, Google AI ModeSingle brand from NOK 880/month (7,500 credits). Excl. VAT. 7-day trial with payment cardReports metrics including Visibility Score, Share of Voice, Sentiment and Citation Rate.

We make Cited ourselves, so read our row with that in mind, and check prices and terms with each vendor, since they change. When you evaluate a tool for measuring AI visibility, here is a good test: does it ask you to trust a percentage, or does it let you open the answers behind it?

You can also get free numbers from the search engines themselves. Google counts traffic from AI Overviews and AI Mode in the regular search data in Search Console. Bing Webmaster Tools has AI Performance, which shows how often your content is cited in Copilot and in AI summaries in Bing. Both measure traffic to and citations of your web pages, not whether your brand name is mentioned. Mention rate is something you have to measure yourself.

How often should you measure AI visibility?

Measure regularly, and read the trend across several measurements instead of each one on its own. By hand, once a month is enough for 30 questions. K—lab measures every other day, and we still read the trend over weeks, not days, because of the variation described above.

After a change, such as a new page or an updated robots.txt, wait a few weeks before drawing conclusions. ChatGPT can only use your pages in search if you let in OAI-SearchBot, and OpenAI says changes to robots.txt can take about 24 hours to take effect.

What do you do with the numbers?

Use the numbers to choose your next action, not to report a percentage. Start with the mention rate on questions without a brand name, and see which questions you are missing from and who is mentioned instead.

Three steps that often follow from the measurement:

  • Find the questions competitors win, and see which sources the answers rely on.
  • Create or improve the page that answers the buyer question, with the answer first. Google writes in its guidance on AI features that ordinary good search engine optimization is what counts.
  • Measure the same questions again, and compare with your noise level.

What counts as a good mention rate depends on the industry. The typical Norwegian B2B company is mentioned in 8% of AI answers about its industry, according to K⁠—⁠lab's measurement of Norwegian B2B companies.

Want to know where you stand? Try the free check to see how often ChatGPT, Copilot and Google mention you. If you want to follow the trend over time, you will find Cited here.

Frequently asked questions about measuring AI visibility

How many questions do I need to measure AI visibility?

At least 20, ideally 30 to 50, without the brand name in the question. K⁠—⁠lab uses a fixed set of buyer questions that are measured every other day.

Can I measure AI visibility for free?

Yes. You can ask questions by hand in ChatGPT and Copilot and log the numerator and denominator in a spreadsheet. You can also take a free check. By hand you get no history, and you have to check yourself that the answer is about the right company.

What is a good mention rate?

It varies by industry. In K⁠—⁠lab's measurement of Norwegian B2B companies, the typical company is mentioned in 8% of the answers about its industry. Compare yourself with your competitors on the same questions, not with a single average.

Is one answer from ChatGPT enough?

No. In under 10% of the compared pairs (November 4 to 18, 2026), the list of mentioned brands changed between two measurements of the same question on the same platform. One answer is a snapshot.

What is the difference between a mention and a citation?

A mention means the brand name appears in the answer. A citation means the answer links to your website as a source. You can be cited without being mentioned, and mentioned without being cited, so measure each one separately.

More weight behind every choice.

Spend less time gathering and explaining results. Spend more time on the actions that actually deserve attention.