AI visibility · the method
AI only names who it already knows
Ask an AI assistant about a service in your field and you get a handful of names. Ask about your own company and you get a tidy answer. That looks the same and it isn't: the first is a market you have to win, the second is a name the machine already had.
For almost every SME, that difference is sharp. Whoever types in their own name gets found, and concludes that things are fine. Whoever asks the question a customer would ask doesn't appear at all. The question on this page is: how do you measure that, and what moves it.
Gepubliceerd · bijgewerkt
The pattern · 28 August 2026
Where those names come from
An AI answer isn't an opinion but a summary of a handful of pages the model trusts at that moment. We counted those pages. In one measurement round covering 40 questions and 4 AI platforms, 1,049 source mentions came back, spread across 583 different sites.
Of those, 415 were mentioned exactly once. So the models graze widely but fall back on a small core: 87 sites came back three times or more, together accounting for 472 of the 1,049 mentions. That's the list that matters, and this is how it's built.
Pages from other agencies on the same topic.
244 vermeldingen · 47 sites
Tools and overviews, often with an interest of their own in the answer.
76 vermeldingen · 15 sites
What people tell each other, not what a company says about itself.
50 vermeldingen · 4 sites
Articles from an editorial team for a trade audience.
43 vermeldingen · 8 sites
The manuals of the models the question was put to.
27 vermeldingen · 6 sites
All sixteen showed up on the three questions that already contained our name.
16 vermeldingen · 2 sites
If in doubt, put it here, not in the group that suits best.
16 vermeldingen · 5 sites
Percentages of the 472 recurring mentions, rounded to whole per cent. The grouping is our judgement, not a measurement; what couldn't be classified with certainty is counted as such and hasn't been added to a group that suited us better.
And this is what it means for a business
Of the 40 questions we asked, 3 carried our own brand name. On precisely those three, our site was cited 16 times. On the remaining 37 questions: 0 times.
So whoever types in their own name gets found. Whoever asks the question their customer asks doesn't. That difference can't be fixed with better copy on your own site, because your own site doesn't appear in that answer at all.
There's a second movement too, on the Google side of the same round: an AI summary already appeared above the regular results for 28 of the 30 measured searches. Whoever isn't in it doesn't exist for a growing share of searchers, even if they're in position four of the blue links.
The method
How we measure that
The pattern above doesn't come from an impression but from a fixed setup: 40 questions, each asked to 4 AI platforms with web search on, plus 30 Google searches including the AI summary above them. That's 160 AI measurements and 30 Google measurements per round. Below is what we do with that, in three steps and five lines.
Step 1 · measuring
The visibility map
The visibility map is the complete list of places where a customer can find you, in three layers: the Google searches with buying intent, the same questions as your customer puts them to ChatGPT, Gemini, Claude or Perplexity, and the external sources that feed those answers.
Every question on the map is actually measured: asked to the four AI platforms with web search on, and looked up in the Google results including AI Overviews. For each answer we record who gets mentioned, at what position, and which sources fed the answer. Alongside the map, this also produces the most important by-catch: the list of sources that feed the answers in your market, ranked by how often AI uses them. That's the working list.
Which sources exist
Eerst de kaart: welke pagina’s, profiles, trade media, comparison articles and communities are about your subject, and which of those are yours. That is an inventory, not a judgement.
Which ones are cited
Then the measurement: for each question on the map, we record which sources fed the answer, per platform and per model. Not what could be found, but what was actually used on the reference date.
Which come up most often
And then the ranking: the same sources, sorted by how often AI uses them. That is the work list, and it is the only list from the scan that says what needs to happen outside your own site. Whether you already appear on such a source is a separate check per source, not an outcome of this count.
Step 2 · summarising
One number: the TAZM score
TAZM stands for Total Addressable Search Market. The TAZM score summarises the whole measurement in one number from 0 to 100: the average of the four components below, where only what has actually been measured counts. In our own baseline measurement that's 2 out of 4, and that composition applies to every number we publish. The score only means something next to that of the strongest player in the same market: the distance is the work.
Google visibility
The share of the measured search queries for which the business ranks in the top 10.
AI visibility
The share of the measured questions for which ChatGPT, Gemini, Claude or Perplexity mentions the business.
Third-party authority
The share of the sources feeding AI answers in which the business appears.
To follow. Not yet measured.
Commercial coverage
The share of the map for which the business has a suitable page of its own.
Measured, with its own reference date. Doesn't count towards our score above.
Step 3 · improving
Own, influence, measure
Improvement happens on three sides at once. What you own: making your own site citable, with commercial pages for every question on the map and technology that lets machines in. What you influence: getting mentioned in the sources that feed the answers, only in the genuine way. Which sources those are comes out of the measurement; whether you're already listed there is a separate check per source. And what you measure: the same measurement again, against the same yardstick, so every euro of work is accounted for against the baseline measurement.
How that cycle works is set out on the AI visibility overview page. One rule is non-negotiable: presence with third parties is never fabricated. No purchased reviews, no self-placed listings. What isn't real works against you in AI answers.
The rules of the game
Five measurement rules
These five apply to every measurement we do, for a client and for ourselves. The pattern at the top of this page is the application of it: read those figures back alongside these rules and you'll see why it says what it says, and why one spot says nothing at all.
- Every question is actually asked and every answer recorded; nothing is estimated or simulated.Our own baseline measurement does not yet fully meet this rule: on one of the four platforms, our measurement runner read the reasoning block instead of the answer, and the source citations were not recorded completely for every platform. That is stated with numerator and denominator on the evidence page, and the fix is included in the next remeasurement.
- Every figure carries a numerator and denominator ("mentioned in 3 of 40") plus a reference date and the measured model name.
- The question list is fixed before the first measurement and is never narrowed silently; every remeasurement uses the same list.If the list does change, it never happens silently: the change gets its own reference date and a new series starts alongside the old one, without a difference between the two. That is the case for our own next round, which is why that round is placed as a new starting point next to the baseline measurement rather than on top of it.
- AI answers differ by day and by user; measurements are samples and are presented as such, never as live data.
- Figures that disappoint are published just as readily as figures that please; our own baseline measurement is public on the evidence page.
And the main promise stays the same: we don't promise positions; we promise a yardstick, and let the before-and-after measurement do the work. Nobody can guarantee a spot in an AI answer, and anyone who does promise that deserves your distrust.
Applied to ourselves
What our own measurement showed
We first applied this method to ourselves, and the outcome isn't favourable: TAZM 7.1 out of 2 of 4 components, reference date 28 August 2026. The pattern at the top of this page is our own source data, and all sixteen mentions of our site fell on the three questions where our name was already known. We publish that figure because a yardstick that only shows favourable outcomes isn't a yardstick.
The full measurement, with numerator, denominator, reference date and the literal platform answers, is on our evidence page, including the answer that is factually incorrect and the platform we didn't read out cleanly.
Frequently asked questions
About the method
What is the visibility map?
The visibility map is the complete list of places where a company can be found: the Google searches with buying intent, the same questions as they are put to ChatGPT, Gemini, Claude and Perplexity, and the external sources that feed those answers. The map differs per company and is measured, not estimated.
What is the TAZM score?
TAZM stands for Total Addressable Search Market. The TAZM score is a single number from 0 to 100 that summarises which share of its own search market a company owns: the average of four components (Google visibility, AI visibility, third-party authority and commercial coverage), where only actually measured components count. In our own baseline measurement of 28 August 2026, that is two of the four, and the score always carries that composition with it.
Why do AI systems cite third parties more often than company websites?
AI assistants build answers from sources they trust: trade media, comparison articles, communities and reviews. In our own measurement of 28 August 2026, sources of various kinds came back for choice questions, and by no means always the business's own site. Anyone who wants to appear in AI answers therefore needs to be present both on their own site and in those sources.
What is source analysis and why is it the most distinguishing part of the scan?
Source analysis is the third question of the measurement: not just whether an AI answer mentions you, but which sources fed that answer. An AI assistant builds its answer from a handful of pages it trusts at that moment; if you are not among them, better text on your own site will not help. We therefore measure, per platform, which sources are cited, which of those are your own and which are third-party, and how often each source recurs. That frequency list is the working list. What that produced for us, per platform and verbatim, is on our evidence page.
Do you publish your own measurements?
Yes. We apply the method to ourselves first and publish every measurement openly on the evidence page, including our baseline measurement of TAZM 7.1 out of 2 of 4 components on 28 August 2026, and the figures we're not proud of. Anyone selling a yardstick should dare to read it themselves.
Same method, applied to your market
The scan is the first step: the map of your market measured, your score with its breakdown included, and an honest verdict on whether we're a fit.