Original research
Original research on how AI systems read the web
Studies built by measuring a stated population directly, then publishing the run that produced every figure. Each page names its frame, its date and its failure count, and links the raw rows.
Every study
No studies are published yet. Everything in this family is held back until it clears the depth gate — sourced, dated, and reviewed by a person.
How these pages are produced
Every study here is generated from a committed run artifact. Nobody types a figure into the page: the measurement runs, the run is committed with its date, and the page is rendered from that file, so the tables, the shares and the counts all regenerate from the artifact rather than being transcribed. A new run replaces the current page and the previous run stays frozen at its own dated URL, so a number cited months ago still resolves to the run it was taken from.
A study only renders while its artifact passes the gate: if the artifact is missing, older than two refresh cycles, or the run's fetch success rate falls under the floor, the page returns 404 instead of showing an aged number. There is no “coming soon” state, which is why this list can name a study whose page is not currently rendering.
The population, the fetch and parse rules, how failures are counted and what each published column means are written down in the methodology.