FINDINGS
Findings
Measurement notes, published as they happen. Not a blog: there is no publishing schedule here, only what we actually found.
- 34.8% of the variance in 'brand mentions in AI answers' is resampling noise
We ran a variance decomposition across 12,933 AI-answer responses. Resampling noise accounted for 34.8% of the variance, against 0.7% for the brand signal being tracked.
- The Cloudflare AI-bot switch nobody checked was even in the path
We checked whether the layer holding the AI-crawler setting was actually in the request path on our own estate. It was not, on any of the domains we checked.
- Nine domains added named AI-crawler blocks. All eleven groups were identical to the default
The advice going round is to add a named User-agent block per AI crawler. We checked what that actually does to robots.txt matching, and on the sites we checked, the naive version of it can open access it was meant to close.
- Six pieces of AI-readiness advice we tested and are telling clients not to bother with
The fix list includes what we're recommending against, with the evidence for each. Four are site-level changes, two are measurement practices worth dropping.