You can only steer what you measure. That also goes for your visibility in AI answers. Yet I often hear: ‘I have no idea whether ChatGPT mentions my brand’. Understandable, because there is no Search Console for AI. But you don’t need an expensive tool to measure it. You need a method.
In this article I share the repeatable method I use myself to measure a brand’s citation share in AI answers. Free, with nothing more than a browser and a spreadsheet. It takes some discipline, but it gives you hard numbers to steer by.
First this: what are you actually measuring?
Before you start, keep two concepts sharp. They are often used interchangeably, but the difference matters.
Mention: your brand or name appears in the answer, without a source link.
Citation: you are named with a real source reference or link to your site. This is the strongest signal and it also drives clicks.
Measure both, but keep them apart. A brand can be mentioned often without ever being cited, and the other way around. Your citation share is the number that matters most for GEO.
And remember: AI answers are not deterministic. The same question does not give the same answer every time. That is why this method revolves around repeating and measuring over time, not a single snapshot. Why AI systems pick what they pick is covered in my article on how AI picks sources.
The method in six steps
Step 1: Define your scope and platforms
First choose what you measure. Usually that is a brand or domain, and the competitors you compare yourself against. Then pick the platforms that matter for your audience. Think ChatGPT, Perplexity, Google AI Overviews and Gemini. Start with just ChatGPT if needed and expand later.
Step 2: Build a fixed prompt set
This is the heart of the method. Put together a fixed list of questions you want to be found for, and reuse it every measurement. Only with a fixed set can you compare results over time. Aim for a mix:
- Informational questions: ‘how does [topic] work?’
- Commercial questions: ‘best [service or product] in 2026’
- Comparisons: ‘[you] versus [competitor]’ or ‘alternatives to [competitor]’
- Brand questions: ‘what is [your brand]?’ and ‘is [your brand] reliable?’
Start with ten to twenty prompts. Enough to get a picture, not so many that you give up. Write them down in your spreadsheet, because this becomes your fixed yardstick.
Step 3: Measure under controlled conditions
To measure fairly, you have to remove noise. Otherwise you measure your own personal history instead of what an average user sees. Stick to these rules:
- Start a new chat for every prompt, so previous answers don’t influence you.
- Turn off personalisation and memory, or measure logged out or in a private window.
- Repeat each prompt a few times, three times for example, and note what you see on average. That captures the variation.
- Measure in the language and region of your audience. Dutch and English answers differ.
- Note the model and the date for every measurement. Models change, so this is your context.
Step 4: Record per answer
Log what you see per prompt. Keep it simple and consistent. These are the columns I use in my spreadsheet:
| Prompt | Platform | Model + date | Mentioned? | Cited + link? | Competitors mentioned |
|---|---|---|---|---|---|
| 'best [service] 2026' | ChatGPT | GPT, 12-06 | Yes | No | competitorA, competitorB |
⬇ Download the tracking sheet as a spreadsheet template (CSV)
Under ‘Competitors mentioned’, note who does appear when you don’t. That is gold, because it shows which sources the model does trust. You will often see a pattern: the same few sites keep showing up.
Step 5: Calculate your citation share
Now turn it into a number. The basis is simple:
Citation share = the number of prompts in which you are cited, divided by the total number of prompts, times one hundred.
If you are cited in 4 out of 20 prompts, your citation share is 20 percent. Do the same for ‘mentioned’ if you also want to track mentions. Want to sharpen it further? Calculate your share relative to competitors: how often you appear, divided by all the times you or a competitor appears. That is your share of voice in AI answers.
Step 6: Repeat and follow the trend
A single measurement is a photo. The value is in the film. Repeat the whole set at a fixed rhythm, monthly for example, with exactly the same prompts and conditions. Put the numbers side by side and watch the movement. Does your citation share rise after a content round? Then your approach works. Is it flat? Then you know you need to do something different.
Common mistakes
- Measuring once and drawing conclusions. Because of the variation in AI answers, a single measurement is almost worthless. Repeat.
- Measuring logged in with your own history. Then you measure your personalisation, not the reality for others.
- Using different prompts every time. Then you can’t compare anything. Keep the set fixed.
- Not noting the model and date. If the numbers move later, you won’t know whether it was you or a model update.
- Only looking at yourself. Who does appear tells you where the opportunities are.
What do you do with the outcome?
Measuring is not a goal in itself. The numbers point the way. Do you see the same competitors being cited again and again? Analyse their content and authority. Are you mentioned but not cited? Work on source-worthiness, meaning clear structure and demonstrable authority. Do you see a category of questions where you are completely absent? That is your first content opportunity.
That is how measuring becomes the foundation of your GEO strategy, instead of a guess. Rather not track this yourself? In a GEO audit I measure your citation share, benchmark you against competitors and deliver a priority list to become more visible. Book a call any time.
Frequently asked questions
Do you need an expensive tool to measure AI citations?
No. With a fixed prompt set, controlled conditions and a spreadsheet you get a long way. Paid tools make it more scalable, but the method also works for free.
Why does ChatGPT give a different answer every time?
AI answers are not deterministic. Models pick words based on probability, and live search results change. That is why you measure repeatedly and look at the average and the trend.
What is a citation share?
The share of your prompts in which you are cited with a source reference, expressed in percent. If you are cited in 4 out of 20 prompts, your citation share is 20 percent.
How often should I measure?
A fixed rhythm works best, monthly for example, with exactly the same prompts and conditions. That way you can see whether your actions have an effect.
Sources and background
This method is based on my own work in GEO audits, combined with what is publicly known about how AI search systems pick sources. For the background on the latter, see my article How ChatGPT picks sources on this blog, and OpenAI’s public documentation on ChatGPT Search.