Does ChatGPT mention you? Here is how to measure it
More and more people search through ChatGPT, Claude, Gemini and Perplexity. You cannot see how often you are mentioned, but you can measure it with a fixed set of questions. This is how.
In Google you can see exactly which keywords you appear for, how often and in what position. Search Console gives you those numbers for free. For ChatGPT and the other AI search engines nothing like that exists. There is no dashboard, no export button and no official data.
That does not mean you are in the dark. You can measure whether you are mentioned, just in a different way: not by asking the source, but by asking the questions yourself and keeping track of the answers.
Why AI search engines pick different sources from Google
Google lists ten results and lets you choose. A language model does something else: it gives one answer and names a handful of sources with it. Where Google offers ten chances to be found, an AI answer has room for two or three.
Which sources those are depends on where the model gets its information. Sometimes that is a search it runs live, sometimes it is what was taken in during training. Either way, pages that answer a question clearly and completely have the advantage, and pages that mostly sell are at a disadvantage.
In practice: a page that ranks well in Google gets mentioned more often, but it is no guarantee. The reverse happens too. We see pages that rank only moderately in Google show up in AI answers, because they explain the question well.
Measuring with a fixed set of questions
The only way to say anything useful about this is a sample. You ask the models a fixed list of questions, the same ones every month, and you record what comes back. Not once, but repeatedly, so you can see a change.
You build that list from the questions your customers really ask. For an installer in Nijmegen, for example:
- Who installs heat pumps in Nijmegen?
- What does a heat pump cost including installation?
- Which installer is good for underfloor heating in Gelderland?
- How much of a grant can I get for a heat pump?
For each question you track three things: are you mentioned, are you cited with a link, and which competitors do appear. The third one is often the most useful. It shows who the models currently treat as the authority.
What you can do with the outcome
If you never appear in the answers and your competitor always does, look at which of their pages gets quoted. Usually it is a piece that answers the question directly, with a clear heading and no padding. That can be matched.
If you are mentioned but without a link, the model is missing a source to point to. That is a sign the information is somewhere on your site, but not on a page that is recognisable as the answer to that question.
And if you do appear in the answers, you now know which pages are doing it. Those pages are worth looking after.
What this measurement is not
We have to be honest here, because a lot is being claimed on this point. A fixed set of questions is a sample, not the full picture. Language models do not give the same answer every time, they differ by country and by user, and they change without notice.
Nobody can tell you how often you are mentioned in ChatGPT, because those numbers simply do not exist outside OpenAI. Anyone who promises them is measuring something else or making it up.
What a sample can do is show a direction and make changes visible. If you are mentioned for ten of fifteen questions where last month it was six, something has improved. That is enough to steer by.
