Tracking AI visibility: are you mentioned and quoted by AI?
You can have it measured whether ChatGPT, Gemini, Perplexity and Claude mention your company and quote it as a source, using the same questions every quarter. Each percentage is accompanied by the number of responses and the margin of error, for example 216 responses in my bank measurement. This is not suitable if your site is hardly found in Google; in that case, start with the audit.
Updated: 19 November 2026
This is the measurement part of my SEO and GEO process, intended for those who want to show a board of directors what is changing. I ask each question multiple times to each assistant. I do not yet measure Google AI Overviews and Copilot.
SEO is being found in Google. GEO (generative engine optimisation) is being mentioned and quoted in the answers of AI assistants.
- Four AI assistants: ChatGPT, Gemini, Perplexity and Claude
- Each question asked multiple times, because an AI answer varies each time
- The number of responses and the margin of error for each percentage
- Distinction between mentioned, quoted and found
- The same measurement with the same questions every quarter
Example from my own bank measurement
216 AI responses to 18 customer questions, measured with ChatGPT, Gemini, Perplexity and Claude.
For banks, 14 out of 216 responses mention no bank at all (6.5%, margin of error approximately 3 percentage points).
What is the difference between mentioned, quoted and found?
AI visibilityAI Visibility How frequently and effectively your business appears in the responses provided by AI assistants. is the share of AI assistantAI Assistant An AI assistant is a chat program such as ChatGPT, Claude, Gemini, or Perplexity that you can ask a question in natural language and receive a written answer. responses in which your company appears. I count that in three ways.
- Mentioned: your name is in the body text of the answer.
- Quoted: your site is listed as a source with the answer.
- Found: the assistant saw your page during the search, but did not use it.
AI visibility is a share and not a yes or no, because an AI assistant answers the same question differently every time. A screenshot shows one answer out of many possible ones. That is why I ask each question multiple times and count across all answers combined.
The difference between found and quoted points to the place to look. If your page is not found, I first look at the search step: indexing, accessibility and the positionPosition Your position is the place your website holds in the search results for a specific keyword. Position 1 is the top organic result. in Google. If it is found but not quoted, I look at the page itself. With Gemini, I cannot measure “found”, because it does not provide that data. What GEOGEO GEO stands for generative engine optimisation: increasing the likelihood that AI assistants such as ChatGPT recognise and mention your business when someone asks them for advice. exactly entails can be found at /geo-optimisation/.
How does a measurement work?
TrackingTracking Recording what visitors do on a website, such as which pages they view and what they click on. AI visibility only works if every measurement is carried out in the same way. I repeat the same steps every quarter.
01
Establish the question set.
I write down the questions once as your customers ask them, without your brand name in them. It is a fixed set of questions, tailored to how your customers search. The set does not change per quarter, otherwise you have nothing to compare.
02
Ask the same questions again.
I ask each question multiple times to each assistant, so that the noise of a single answer is averaged out.
03
Count who is mentioned, quoted and found.
Per assistant, you see your own share, with the number of responses and the margin of errorMargin of Error The degree to which a measured figure may deviate from reality. included.
04
Compare with the previous measurement.
After the first measurement, you have a baseline. From the second measurement onwards, you can see what has changed since the previous one.
What does a measurement look like, with the number of responses and margin of error?
A percentage only means something with two additional things. The sample is the number of responses on which the percentage is based. The margin of error is how much the percentage can deviate by chance. The more responses, the smaller the margin. In the worst case, with a confidence level of 95%, the margin is approximately 18 percentage pointsPercentage Point The difference between two percentages, calculated by simple subtraction. for 30 responses, 10 for 100 and 5 for 400 (own calculation).
An example from my own bank measurement: 216 AI responsesAI Answer The generated response provided by an AI assistant in answer to a query. to 18 customer questions, measured with ChatGPT, Gemini, PerplexityPerplexity An AI search engine that provides answers with numbered citations. and Claude. For banks, 14 out of 216 responses mention no bank at all (6.5%, margin of error approximately 3 percentage points). Per individual question, there are 12 responses, and then the margin in the worst case is approximately 28 percentage points. Such a figure per question is an indication and not a finding. Read the full measurement at /which-bank-does-the-ai-mention/.
A difference between two measurements that remains within the margin, I call noise. One measurement is an indication, a series of measurements is a trend. What is disappointing or remains within the noise, I also show.
What objections does your board have, and when does tracking not yet make sense?
An AI answers differently every time, so what does a measurement say?
A measurement says something about a share and not about a single answer. I ask the same question multiple times, count across all answers and include the margin of error. If a new measurement differs less than that margin from the previous one, I do not call that a change.
Can I not just ask this in ChatGPT myself?
You can, and it is a good start. But your own question gives one answer, which can differ per user and per time. A screenshot from a colleague therefore shows a different result than yours. My measurement uses fixed questions without your brand name, on the same assistants, always in the same way. This allows you to justify a difference between two quarters.
When does tracking not yet make sense?
AI assistants also search for themselves. If your site is still hardly findable in Google, then improving first is more useful than tracking. In that case, start with the SEO and GEO audit or the SEO and GEO process.
Which service fits your question?
SEO and GEO audit
Choose this if you first want to know why you are not being mentioned. One-off, with a plan.
Keyword research
Choose this if you want to know which questions and words your customers use. Those questions form the question set for this measurement.
SEO and GEO
Choose this if you want to improve how you are found in Google and mentioned in AI answers. The measurement on this page is the measurement part of that process. The process and the price are on that page.
AI swarm for discoverability
Choose this if you want to know how I do the work. It is a method and not a product.
How to choose: if you want to know why, start with the audit. If you want to improve, that is SEO and GEO. If you want to track progress, you can read here how that measurement works.
Frequently asked questions
Do you measure by country or language?
Short answer: by default I measure for the Netherlands; another language or another country is measured separately.
How do you measure when an AI does not give the same answer every time?
Short answer: by asking each question multiple times and calculating a percentage across all answers combined.
Each percentage is accompanied by the number of responses and the margin of error. A difference within that margin is what I call noise.
How often do you measure, and when do I see whether anything is changing?
Short answer: every quarter, and the first comparison follows after the second measurement.
After the first measurement you have a baseline. After that, you see what has changed each quarter, including the number of responses and the margin of error.
What does tracking AI visibility cost?
Short answer: tracking AI visibility does not have a separate fee; it is the measurement component of the SEO and GEO package.
The pricing for that service can be found at /seo-en-geo/.
What exactly do you measure in an AI answer?
Short answer: whether you are mentioned, whether you are cited as a source, and whether your page was found.
Mentioned means your name appears in the text, cited means your website is referenced as a source, and found means your page appears in the assistant’s search resultsSearch Results The page containing links and advertisements that you see after searching for something on Google.. I do not measure how positive an answer sounds about you.
What happens if I am not mentioned anywhere?
Short answer: the initial audit then serves as your baseline, and I first establish where things are going wrong.
If your page is not found, I examine its discoverability. If it is found but not cited, I look at the page itself. Which improvements take priority after that is something I detail in the SEO and GEO project.
Which AI assistants do you measure?
Short answer: ChatGPT, Gemini, Perplexity and Claude.
Each question in the fixed set is sent to all four, multiple times.
How do you start with tracking AI visibility?
You start with the free website scan. You answer ten questions about your company and your customers, and one of them is about ChatGPT. Within two working days, a plan with points for improvement follows, reviewed by a human. The scan is an initial look and not yet the full measurement with a fixed set of questions.
Want to read what the process entails first? View SEO and GEO
