Method
How we measure visibility in AI answers
A measurement is only worth as much as the explanation of how it was made. This page describes our method completely – including what it cannot do.
In short: we ask ChatGPT, Gemini, Perplexity, Grok weekly the questions a customer would ask – without naming the company – and count in how many answers the company is mentioned. Every answer is stored. Since February 2026: 38,917 answers to 4,781 questions in 139 measurement runs.
Four systems, weekly
We measure in the four AI systems that customers in Germany actually use for provider questions. Each system is queried via its official interface with the standard model available there; the answer is stored unchanged with time stamp, model and question.
- ChatGPT
- Gemini
- Perplexity
- Grok
Two further models (Claude and GPT-5 (Vollversion)) were part of the measurement until May 2026 and are paused since then. As long as they are paused, we say "four" – not six.
How the questions are built
No company name in the question
Whoever asks "What does company X offer?" only measures that the model can read. We ask the way an interested party asks who does not know the company yet: "Who installs photovoltaics with storage in my region?"
Real customer intent
The questions cover the situations in which someone looks for a provider: buying decision, comparison, local search, problem and solution, service, reputation. Typically 100 questions per company, developed together in the first call.
The same questions every time
A question stays unchanged across all measurement runs. Only then is a change in the number a change in visibility – and not a change in the question.
In German
We measure the way German-speaking customers ask. Other languages and countries on request.
How we count
- Mention
- A mention counts when the company is named in the answer text – by name or by its website. We record the position (first, among several, at the end) and the context (recommended, only listed, critical).
- Mention rate
- Share of answers in which the company is mentioned, in percent – per AI system and overall. This is the one number we work with.
- Competitors
- Every other provider named in the same answer is recorded. So you do not only see your number, but who is named instead of you.
- Sources
- Where a system cites sources, we store them – and whether your own website is among them.
- Repetition
- AI models do not answer deterministically. That is why we measure repeatedly and look at trends, not at a single value.
What the measurement cannot do
- It does not guarantee a mention. The models decide for themselves what they name – nobody can buy a placement, and we do not promise one.
- It does not measure how many people actually asked this question. It measures what the systems answer when the question is asked.
- It cannot look into the training data. It shows the effect, not the cause – the cause we derive from the technical audit and the content.
- A single measurement is a snapshot. Only the comparison across several runs shows movement.
- It measures the standard answers of the systems, not personalised answers of a logged-in user.
“We do not guarantee a mention. The models decide for themselves. What we guarantee is a clean measurement before and after – and that we tell you what we found, even if it is inconvenient.”
The current measurement stock
38,917
AI answers stored
4,781
different questions
139
measurement runs
16.5 %
of all answers name the measured company
As of 15/09/2026, own database. Companies are not named.
Questions about the method
Why do you not ask for the company name?
Because the answer would only show that the model knows the name. Meaningful is only whether the model names the company on its own when someone asks for a provider – without the name in the question.
How often do you measure?
Weekly. For the analysis, one full run with all questions; for ongoing care, every month a report with the development over time.
Which AI systems do you measure?
ChatGPT, Gemini, Perplexity, Grok – four systems, via their official interfaces. Two further models are paused at the moment; as long as they are, we say four.
Can you guarantee that my company will be named?
No. Nobody can. The models decide for themselves. We can measure where you stand, do the work that makes a mention more likely, and measure again.
Are the numbers on this website real?
Yes. Every number on this site comes from our own database and carries a date. As of 15/09/2026. We do not name companies without their consent – that is why the case examples are anonymised.