Research
Original, in-depth studies on how AI search actually works. Each one starts with a real question, runs a method we can defend, and ends with findings you can quote and cite.
We publish our method and predictions before we look at the data.
That means you can read what we expected to find, then check it against what we actually found. A study we got wrong stays up, and we lead with it.
Pre-registered studies
1 study open
These are running now. The question, the prediction and the full method are already public. The results aren't in yet, and they get published either way.
Pre-registration
Does anything actually read llms.txt?Question: In normal operation, do AI systems fetch llms.txt at all?
The llms.txt standard has been adopted fast, but nobody has convincingly shown that AI systems actually fetch the file. We're logging requests at the edge and comparing llms.txt fetches against robots.txt and sitemap.xml on sites that all serve a valid file.
Registered 20 July 2026, results expected 30 September 2026
Read the protocol
Published studies
None published yet
No results published yet
The studies above are still running. Results land here when their windows close, and everyone on the list gets them by email the same day.
Get notifiedBe first to read new Tilio research
One email when a study is registered, and one when its results land. No digests, no noise.