You can measure AI search visibility, but only if your measurements are comparable. That works if you lock in three things upfront: which questions you measure, what exactly you count and which AI system you measure in. Then you get a baseline you can run again next month, without any debate afterwards.
No tricks, just a way of measuring that you can later compare with earlier measurements.
1) Start with fixed questions (so you’re comparing like with like)
Choose a small set of questions you’ll keep using. Write the questions out exactly (copy and paste), so you measure the same input every time. For each question, also record:
- The exact text (same words, same spelling).
- Whether it includes a place name/region (and which).
- The question type (e.g., explanation, local, recommendation, costs, comparison).
- Mention: is your company name in the answer to a question where your name was not included?
- Source citation: is your domain named as a source/citation?
- Google AI impressions: do you see impressions from generative AI features in Search Console?
- Measurement date (the date you did the measurement).
- Exact question (copy/paste).
- AI system (e.g., ChatGPT or Perplexity).
- Primary metric (e.g., “mention: yes/no”).
- Sources/domains (optional, only if you want to compare them again later).
- Choose 5 fixed questions (only question types that truly fit your offering).
- Measure them in 2 fixed AI systems (and keep that set the same).
- For each question, count only your primary metric (e.g., “mention: yes/no”).
- Repeat on a fixed cadence (e.g., monthly) and note the measurement moment.
- Note which changes went live on your website and when; treat that as context, not as a proven cause.
- Ai optimization guide (developers.google.com)
- Ai features (developers.google.com)
Why this helps: in Klusio’s AI research (2026-Q3) we explicitly define the measurement baseline and counting rules, precisely because small differences in question wording can change your outcome. That’s not proof of causality, but it is exactly why you should keep your measurement questions stable.
2) Agree on counting rules: what does “AI search visibility” mean for you?
Measuring goes wrong if one time you count “a source link”, another time “a mention”, and sometimes “it felt like it went well”. Those are three different outcomes.
So choose one primary metric and make it strict. These are three options you can define clearly:
In our reporting, we use the counting rule “mentioned”: the name of the tracked company appears in the answer to a question without that name. That definition (and the scope) is on the reporting page.
3) Measure per AI system separately (don’t invent an ‘AI ranking’)
Measurement results from ChatGPT, Perplexity, Gemini, Claude and Google’s AI features are not automatically the same. So: create a baseline per system, with the same questions and the same counting rules.
In our AI research (2026-Q3) you’ll also see why: overlap between systems is limited, and a mention is not a fixed position. Positions in an AI answer also aren’t stable enough to count as a “ranking”.
4) Record each measurement so you can repeat it later
If you want to compare later, you need to write down the result so someone else can verify it too. For each measurement, record at least:
5) Want to measure Google AI? Use Search Console (with a separate AI report)
For Google, there’s an official place to measure. Google indicates you can measure content performance in generative AI features on Google Search and Discover with the Generative AI performance report in Search Console1.
In addition, sites that appear as a link in Google AI features (such as AI Overviews and AI Mode) are also included in the total Search Console data in the Performance report (search type “Web”). This is stated in the documentation on AI features and your website2.
Keep those two reports separate in your measurement: AI impressions are a separate report and you don’t simply add them to your regular “Web” performance.
A concrete example of that separation is in our case study of Merkenbureau Van Asselt, where Search Console shows both impressions/clicks and separate AI impressions, with separate measurement windows.
And when evaluating Search Console data: Google recommends focusing more often in analysis on trends in impressions and clicks rather than only on average position.
6) A simple measurement routine you can repeat every month
Want to keep it practical, but still tight enough to compare? Then this works well:
Would you rather not build this in a spreadsheet? On our page about measuring AI visibility we show how we capture this in our measurements.
Finally: measuring is the starting point, not the conclusion
Once your baseline is in place, you can see: are you being mentioned more often, in which systems, and for which question types. Only then can you make a clear-headed choice about what the best next improvement is.
Mainly unsure about tooling (measurement, writing, or implementation)? Then this piece on how to choose an SEO tool for measuring, writing, or work that goes live will help.
