Guide · Measurement
How to measure AI visibility (also in AI Overviews)
By Kristian Stig Henriksen · Dear Future
You measure AI visibility by asking the questions your buyers ask, directly to the assistants, and recording four things in every answer: is your brand named, is it cited with a link, in what tone, and how large a share of the answers belongs to you versus your competitors. Done once, it is a snapshot; done daily, it is a curve you can act on. Here is the full approach, including how to do it yourself.
The four numbers that make up AI visibility
- Mention: is your brand named in the answer? It counts even when the link is missing; buyers remember names, not URLs.
- Citation: does the answer link to your domain as a source? Citations are the answer's footnotes and the closest thing to a click.
- Tone: are you described positively, neutrally or negatively? Being named as a warning is not visibility you want.
- Share of voice: your share of all brand mentions in the answers across the category's questions. It is the number that makes the trend comparable over time and against competitors.
How to measure it yourself, for free
Write ten to twenty questions buyers in your category actually ask, without your brand name in the question. Ask them in ChatGPT and Gemini with web search switched on, from the country your customers sit in, and note for every answer: who is named, who is cited, in what tone. Repeat on fixed days, and store the answers so you can check back.
The method is free and real, but it has three limits you should know. Answers vary from run to run, so few measurements give noisy numbers. It is hard to keep conditions constant by hand: same model version, same country, same search access. And from around a hundred questions times several assistants times daily runs, it becomes a full-time job. That is exactly the work measurement tools automate.
Measuring and optimizing in Google AI Overviews
AI Overviews are Google's AI-generated summaries above the search results, launched broadly in 2024. They have started appearing on more Danish searches, and they behave like the rest of the field: a synthesis with a few cited sources. The measurement is the same: do your important searches trigger an overview, are you named in it, and are you cited as a source?
AI Overviews optimization is not a separate discipline. The overview is built from pages Google has already crawled and ranked, so what works is the same as in the rest of GEO: a direct answer at the top of the page, questions as headings, and structured data matching the visible content. The difference is that your SEO foundation matters more here, because the sources are drawn from the index.
When is a change real?
Any share measured on a sample fluctuates. Before you celebrate a jump from 30% to 34%, you need to know whether it is a development or noise. We therefore show every estimate with a 95% confidence interval using Wilson's method, which holds up when the sample is small or the share sits near 0% or 100%. The rule of thumb: a move only counts as a development once it stays outside the normal variation for several weeks.
Also read what GEO is and the guide to Share of Voice in AI, or see the full method behind our numbers.
Skip the handwork
Get the measurement set up for your brand
We set up a tracked workspace for your category: your questions, your competitors, measured every day with the uncertainty visible.
Get the analysis →Frequently asked questions
How often should you measure AI visibility?
Daily, if you want to see developments and react to them. The models are updated continuously and answers vary day to day, so a monthly sample cannot separate development from noise. Measuring by hand, a fixed weekly rhythm is the minimum; a daily automated measure is the standard we work to ourselves.
What is share of voice in AI?
Your share of all brand mentions in the AI answers on a fixed pool of category questions. If 100 brands are named across the answers and 30 of them are yours, you have 30% share of voice. The number makes your visibility comparable over time and against competitors, regardless of how many questions you measure on.
Can you measure AI visibility for free?
Yes, at small scale: ask ten to twenty buyer questions with web search switched on, note mentions, citations and tone, and repeat on fixed days. That gives an honest snapshot. The limit is scale and consistency: daily measurements across assistants, countries and model versions require automation.