What is evidenced — and what is not
In this field a lot is claimed and little is measured. These pages only carry statements for which a survey exists — with source, sample size and date. Where we know nothing, that is stated too.
- ~5 %
Where AI really cites from
Only around 5 % of AI citations lead to your own domain. 38 % of German ChatGPT source references come from Google Maps.
Which crawler decides
Documented AI crawlers separated by purpose. Why blocking GPTBot costs nothing and blocking OAI-SearchBot makes you invisible.
- 3 / 54
What of AEO really works
In the only controlled comparison, 3 of 54 methods were significant — for question-answer tasks, none.
- 0 / 8
The render gap
Claude does not execute JavaScript. If your text only appears in the browser, the page is empty for Claude.
- 97 %
llms.txt: never fetched
For 97 % of 137,210 domains examined, no crawler ever fetched the file. No provider supports the format.
How we handle evidence
Not every figure carries the same weight. A controlled comparison says more than a correlation, and an agency study more than a claim without data. That is why we note for every figure where it comes from and how sound it is — even when that weakens our own point.
For some questions there simply is no sound answer. Google AI Overviews, for instance, cannot be measured reproducibly from the outside. That then stands as a gap, not as an estimate.