Measurement · Do it yourself
How to check your AI visibility yourself, and what a full analysis adds
By Kristian Stig Henriksen · Dear Future
You can check your AI visibility yourself in an afternoon: write 20 questions your customers might ask, put them to ChatGPT and Gemini with search switched on, and note for every answer whether you are mentioned, whether you are cited, and what the tone is. Repeat for seven days. That gives an honest first picture. It does not give a number you can act on, and the difference is explained at the end.
Why a single check is not enough
The first thing most people do is ask ChatGPT about their own brand. That says almost nothing. Ask about yourself and you get an answer about yourself; customers do not ask about you, they ask about a need. And one answer is a single draw: the same question asked again tomorrow gives a different answer, with different names. So the measurement has to be independent of your brand and repeated over time.
Step 1: Write 20 questions a buyer would ask
Start from what customers ask before they know you: who can help with a particular task in Denmark, what does it typically cost, what is the difference between two approaches, which tools exist. Spread them across the buying journey: some early and explanatory, some comparative, some close to the decision. None of them may mention your name. Write them the way people write, in full sentences and in the customer's language.
Step 2: Put them to two assistants with search switched on
Use ChatGPT and Gemini, both with web search enabled, so the answers build on current pages and can cite sources. Use a new, empty chat for every question so the previous conversation does not colour the answer, and use a profile with no history about your brand. Copy the whole answer including the list of sources.
Step 3: Note three things per answer
- Mentioned: is your name in the answer? Yes or no.
- Cited: does the answer link to one of your pages? Yes or no, and which page.
- Tone: are you recommended, mentioned neutrally, or advised against?
Also note who else is mentioned and cited. That is the competitive picture, and it is often the most surprising part: the assistants rarely recommend the companies you consider competitors, and often someone you have never heard of.
Step 4: Repeat for seven days
Same 20 questions, same two assistants, every day for a week. That gives 280 answers. Only now can you compute shares: what share of answers mention you, what share cite you, and how often the tone is positive. Our guide to the confidence interval explains why 280 answers give an interval rather than a precise number, and how wide it is.
Step 5: Read the result as a map, not as a grade
Split the questions into three groups: those where you are mentioned, those where competitors are mentioned and you are not, and those where nobody is mentioned. The middle group is your work list: the answer already exists there, and it points to someone else. Look at which pages are cited in those answers; that is the kind of page you are missing. The last group is empty seats, where a good page can become the source.
What a full analysis adds
The do-it-yourself method gives you the picture. It does not give you the movement, and movement is what you act on. A full analysis does four things the spreadsheet cannot:
- Scale: 120 questions instead of 20, designed as a market survey across topics and buying stages, so the number covers the category and not your assumptions.
- Daily measurement, every day, without anyone having to remember, so a change is visible when it happens and can be read against the week before.
- The competitors on the same questions, under the same rules, so share of voice is a share of the whole answer and not just your own number.
- The uncertainty on every number, a control group of questions that are left alone, and a roadmap where every recommendation is tied to the questions it is meant to move.
Read more about what it contains on the AI visibility analysis page, and about how the number is made in the method.
Read on: how to measure AI visibility, the confidence interval behind the number and what the analysis contains.
From spreadsheet to measurement
Want the number every day?
Get an AI visibility analysis: 120 questions, two assistants, daily measurement and a roadmap that points at the questions that can be moved.
Get the analysis →Frequently asked questions
Are there simple AI visibility checks without any calculation?
Yes. The simplest is to put 20 customer questions to ChatGPT and Gemini with search switched on and note whether you are mentioned and cited. It needs no calculation and gives an honest first picture. Shares and uncertainty only arrive when you repeat it over several days.
Can you check a brand's AI visibility before investing in a full analysis?
Yes, and you should. The method in this guide takes an afternoon to set up and a week to run. It shows whether the assistants mention you, and who else they mention. A full analysis is the answer when you need to measure movement over time and across competitors.
Are there demos or free access to tools that check AI visibility?
Several platforms offer trial periods, and most require you to define the questions yourself. Whatever the tool, the same rules apply: the questions must not mention your brand, search must be switched on, and one answer is not a number. The method in this guide works without any tool.
How many questions do you need to test before the number means anything?
For a first picture, 20 questions asked daily for a week is enough to see whether you are present. For a number that can be compared over time you need more: at 240 answers a day it takes 12 to 13 days to get down to plus or minus half a percentage point around a share of 2%. That is why a full analysis reads in two-week windows.