Which AI engines does ceeme measure today?
Nine, and we ask them the question your customer would ask — with the full answer kept, not just the score. Where the quality of a measurement differs per engine, that shows in your report at that engine and not in a footnote.
-
ChatGPT
OpenAI
searches the live web
-
Google AI Overviews
searches the live web
-
Gemini
Google
searches the live web
-
Perplexity
searches the live web
-
Claude
Anthropic
searches the live web
-
Grok
xAI
searches the live web
-
Meta AI
Llama
answers from its model
-
Mistral
answers from its model
-
DeepSeek
answers from its model
Is it acceptable to count an engine you cannot query?
Counting yes, measuring no — and the difference between those two is why this section exists. Copilot holds almost seven per cent of assistant use in Belgium. Microsoft offers no public endpoint to ask what Copilot answered someone; what does exist is the underlying model through a cloud platform, and that is a different thing from the assistant your customer uses.
What is the difference between measuring more engines and measuring the right ones?
Measuring more engines produces a longer report; measuring the right ones changes your decision. A provider ticking thirteen engines in Belgium is measuring nine that together reach under one per cent there — and the question of whether you get found in ChatGPT is not answered a millimetre better for it.
What share of my market does the measurement genuinely cover?
For six markets we have real country figures, below, with source and month. For any other market the weighting falls back on the world picture, and your report says so in as many words rather than pretending. That is not modesty: a weighted score with an invented weight is worth less than an unweighted one, because you cannot check it.
| market | engines with a measured share | source and month |
|---|---|---|
| Belgium | ChatGPT 73.8% · Gemini 10.4% · Perplexity 4.5% · Claude 4.2% | Statcounter Global Stats, AI Chatbot Market Share Belgium — July 2026 |
| Japan | ChatGPT 58.7% · Gemini 21.2% | Statcounter Global Stats — July 2026 |
| China | ChatGPT 63.1% · Gemini 27.4% · DeepSeek 3.8% | Statcounter Global Stats — July 2026 |
| France | ChatGPT 72.4% · Gemini 9.1% | Statcounter Global Stats — July 2026 |
| Germany | ChatGPT 67.8% · Perplexity 10.7% | Statcounter Global Stats — July 2026 |
| Russia | Perplexity 48.1% · ChatGPT 30% | Statcounter Global Stats — July 2026 |
Share of AI assistant use per country. Where no country figure exists, your report weighs with the world picture and says so.
What does an extra engine cost in the measurement?
In compute time: one extra full round of questions, because every question is put to every engine separately and the full answer is kept. In money it depends on your route, and that sits on one page. What an extra engine does not cost is accuracy: every engine is measured separately and only then weighted, so adding one does not dilute your existing figures.
Frequently asked questions
Why is Copilot not in the list?
Because there is no way to ask what Copilot answered someone. What does exist is the underlying model through a cloud platform, and that is not the same assistant — an answer from that detour would be about a different product from the one your customer is in. The share does stay in the denominator of your coverage.
Do you also measure Google AI Overviews?
Yes, through an intermediary service, and it is marked as a detour in your report. There is no official entrance, so what we measure is what such a service gets back and not what Google tells us directly. Naming that difference is why you can believe that row.
What if my customers are in a country not listed here?
Then we measure there with the world weighting and your report says it is a world weighting. We do not fill in a country figure we do not have; that would be a weight nobody can check, at exactly the spot this whole product is about.
Read on
Free, no account and no card.