How to Measure AI Visibility: Question Design, Sampling and Tracking
To measure your brand's AI visibility, ask ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews a fixed set of questions that do not include your brand name, and record whether you are mentioned, where, how, in what tone, and whether your site is cited. Ask each question at least 3 times, and compare before and after about 4 weeks after publishing.
Also available in：繁體中文
To measure AI visibility, ask ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews a fixed set of questions that do not include your brand name, on a regular schedule, and record for each answer whether you are mentioned, where and how, the tone, and whether your site is among the cited sources. AI visibility is the degree to which a brand is mentioned, recommended and cited in AI answers; ask each question at least 3 times, and measure again about 4 weeks after publishing to tell whether a change worked.
Measurement is both the first and the last step of the GEO loop: before you start, it shows the gaps; after you publish, it verifies the results. This guide covers how to design questions, how many times to ask, which signals to record, and how to track over time.
Why measure AI visibility?
AI answers decide which brands people see. When a Google search shows an AI summary, users click a traditional search result 8% of the time, versus 15% without one (Pew Research Center, July 2025). Rankings and traffic reports cannot show this; you have to look at the AI answers themselves.
Without measurement, you cannot tell which topics to write first, or whether what you wrote made a difference.
First: design your monitoring questions
Monitoring questions are the fixed questions you ask AI on a schedule. Good ones meet two conditions: customers actually ask them, and they do not include your brand name.
If the question names your brand, AI will obviously mention you, and you cannot measure your visibility when nobody prompts for you. Cover five question types to see in which situations you appear:
| Type | Purpose | Example (a bookkeeping firm) |
|---|---|---|
| Core service | Asks directly about what you offer | Which bookkeeping services in Taipei do you recommend? |
| Pain point | A specific problem customers face | I just started a company and nobody knows how to file taxes. What should I do? |
| Comparison | A trade-off between two options | Is it cheaper to do my own bookkeeping or outsource it? |
| Details | Processes, numbers, rules | How much does bookkeeping cost per month for a small company? |
| Beginner | A basic question for newcomers | What does company bookkeeping include? |
Five to 10 questions are enough to start. Once set, do not change them casually: fixed questions are what make results comparable over time.
Which AI tools, and how many times per question?
Cover at least the AI tools your customers use most: ChatGPT, Google AI Overviews, Gemini, Perplexity and Claude. Turn on web search when testing, because that is how people actually use them, and it is the only way to see which pages AI cites.
Each platform documents its retrieval: ChatGPT search indexes pages with OAI-SearchBot (OpenAI crawler documentation), Google AI Overviews cites pages Google has indexed (Google Search Central), and Perplexity searches the web and cites sources in every answer (Perplexity documentation).
AI answers vary slightly each time. Being mentioned once does not mean you always will be; ask each question at least 3 times and count how many of the 3 answers mention you. This mention rate is far more reliable than a single yes or no.
| Times per question | What you can see | Best for |
|---|---|---|
| 1 | Only yes or no, with a lot of noise | A first rough look |
| 3 | Four mention-rate levels: 0%, 33%, 67%, 100% | Weekly routine measurement |
| 5 or more | Finer mention rates, more stable week to week | Key questions and numbers you report externally |
Which signals should you record for each answer?
Each answer breaks down into five signals. Recording only "mentioned or not" loses a lot: a passing mention at the end and a clear recommendation at the top affect readers very differently.
| Signal | What to look for | Why it matters |
|---|---|---|
| Mentioned | Whether your brand name or a common alias appears | The most basic visibility |
| Position | At the start, in the middle or at the end | Readers usually only look at the first few options |
| Type of mention | A clear recommendation, one option among several, or a passing mention | Recommendations persuade far more than passing mentions |
| Tone | Positive, neutral, or with warnings or drawbacks | Negative descriptions cancel out exposure |
| Cited sources | Whether your own site is among the source links | Your page is being used as evidence for the answer |
Watch out for short brand names. A two-letter name is easily mistaken for part of another word, so also record common aliases or full names and match them together.
How do you compare with competitors?
In the same answers, record whether your competitors are mentioned. Share of voice is the number of times you are mentioned across all answers, divided by the total mentions of you and your competitors.
For example, ask 10 questions 3 times each for 30 answers. You are mentioned 12 times, competitor A 18 times and competitor B 10 times, so your share of voice is 12 ÷ (12 + 18 + 10) = 30%.
Share of voice reflects your competitive position better than your own mention count alone: when AI rarely mentions any brand on a topic, 3 mentions may already put you first.
Two situations to look for in your results
After one round, find these two kinds of questions first. They point most directly to what to do next:
- Nobody is mentioned: neither you nor your competitors appear. This is open ground where one good article could become the AI's main source, so write these first.
- Mentioned but not cited: AI recognizes you, but the sources it lists are other people's pages. Your page was not found by AI search or was not citable enough, so check that your site allows AI crawlers and rewrite the content.
How do you track after publishing?
After publishing, ask the same questions again and compare before and after. This is the only step in the GEO loop that proves a change worked.
AI search has to index a new page before it can cite it, so you usually see no change on publishing day. Keep measuring weekly and compare before and after at about 4 weeks; a single week can mislead you because AI answers vary randomly, so trust the trend over any single point.
| Signal | What success looks like |
|---|---|
| Cited sources | Your new article appears among the sources AI lists |
| Mention rate | Out of 3 asks, more answers mention you |
| Position and type | A passing mention at the end becomes a recommendation at the top |
| Share of voice | Your share among you and your competitors rises |
Of the four, cited sources is the most direct: it means AI actually read your article and used it as evidence for its answer.
What if nothing improves?
If nothing changes after about 4 weeks, rule things out from basic to advanced:
- Not indexed: search for the article title in a regular search engine. If it is not there, check that robots.txt allows search and AI crawlers, and submit a sitemap.
- Topic mismatch: do the title and opening directly answer the question you are measuring? If not, rewrite them.
- Not citable enough: does the opening answer directly? Are there specific numbers and verifiable sources? Research shows that adding sources, statistics and quotations improves visibility by roughly 30–40% (GEO paper).
- Too much competition: look at which sites AI cites. If they are all large sites, switch to a narrower topic.
Once an article starts getting cited, compare it with your other articles and apply the difference to the next one. Keep cited articles updated with current numbers and rules so they are not replaced once they go stale.
Frequently Asked Questions
Q: What is AI visibility?
AI visibility is the degree to which a brand is mentioned, recommended and cited in answers from AI tools such as ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews.
It differs from search ranking: ranking is about your position in a list of links, while AI visibility is about whether the AI's written answer includes you and lists your page as a source.
Q: How do I measure my brand's AI visibility?
Ask several AI tools a fixed set of questions without your brand name on a regular schedule, and record five signals for each answer.
The five signals are: mentioned, position, type of mention, tone and cited sources. Turn on web search when testing, and record competitors at the same time.
Q: Why can't monitoring questions include my brand name?
Because if the question names your brand, AI will mention you, and the result says nothing about your real visibility.
People asking AI usually do not know or name a brand. Measurement should simulate that unprompted situation to show whether AI recommends you on its own.
Q: How many times should I ask AI the same question?
At least 3 times, and 5 or more for important questions.
AI answers vary slightly each time. Three asks give you a mention rate, such as 67% when 2 of 3 answers mention you, which is far more reliable than a single yes or no.
Q: Which AI platforms should I measure?
Cover at least the ones your customers use most among ChatGPT, Google AI Overviews, Gemini, Perplexity and Claude.
Each platform retrieves differently, and the same question can produce very different results. Turn on web search when testing so you can see which pages AI cites.
Q: How is share of voice calculated?
Share of voice is the number of times you are mentioned, divided by the total mentions of you and your competitors.
For example, if you are mentioned 12 times and two competitors 28 times combined, your share of voice is 12 ÷ 40 = 30%. It reflects competitive position better than your own count alone.
Q: How often should I measure?
Weekly or every two weeks, using the same set of questions.
A fixed schedule is what reveals trends. After publishing new content, compare before and after at about 4 weeks, because AI has to index a new page before it can cite it.
Q: What does it mean when AI mentions my brand but does not cite my site?
It means AI recognizes your brand, but its answer is based on someone else's pages.
The usual causes are that AI search did not find your page or the content is not citable enough. Check that robots.txt allows AI crawlers and that the page is indexed, then rewrite the opening to answer directly and add numbers and sources.
Start monitoring your brand's AI visibility
Free sign-up, 5-minute setup, an automatic BVS report every week
Start free