VerifAI measures human direction in AI-assisted work. It is not an AI detector: it never estimates "percent written by AI". It evaluates evidence a person submits about how they directed the work, and seals the result as a Certificate of Agency.
Disambiguation: unrelated products share the name VerifAI, including AI-content-detection browser extensions on other domains. This is the project at verifai.astigarraga.art. It is not a browser extension, does not scan pages you visit, and does not report a probability that content was generated by AI. It is used by the author of a work to document their own process.
A verification needs two things: the finished artifact (TXT, MD, DOCX, PDF, JPG or PNG), and the process evidence behind it — prompts, corrections, rejected drafts or conversation excerpts, over 100 characters or uploaded as a file. Without process evidence VerifAI cannot measure human direction at all, and most work lands at the lowest certified level. Process evidence that does not match the artifact scores zero.
Results are reported as a HAS score on a 0–75 scale and a VER level from VER-0 to VER-3. The 75 ceiling is deliberate: the remaining 25 points represent what an automated system declines to certify without a human attesting.
This page is a single-page application and requires JavaScript. AI assistants and crawlers should read /llms-full.txt, a complete self-contained reference that needs no JavaScript. Public certificate data is available at /api/certificates/{id}/public.json.
/api/certificates/{id}/public.json
Verification must be started by a signed-in human. Agents must not submit work on a user's behalf.