Open method, per-signal breakdown, runs on our hardware for free. No verdicts: this measures tell density, not authorship.
Is: a fast, transparent stylometric scanner. It flags patterns that read machine-written — uniform sentence rhythm, AI-favoured phrasing, missing contractions, a too-predictable word stream — and shows you each one, with the measurement behind it.
Isn't: proof of anything. It has false positives on formal, non-native, and heavily edited human writing, and false negatives on lightly edited AI text. That's true of every text detector, whatever their landing pages claim — which is why every result here shows which signals fired, so a score can be understood and contested, not just believed. How to read the score →
ScanForAI runs the open-source aiscan
engine. Tier 1 is a set of open stylometric heuristics (you can read
the exact formulas). Tier 2 runs a small open
language model (GPT-2) on our own hardware to measure how predictable your
text is, token by token — machine text is eerily predictable. No third-party AI
APIs are involved, and anonymous scans are never stored.
The scanner is an open HTTP API — pipe any file straight in:
curl -s --data-binary @draft.txt "https://scanforai.com/api/scan?format=text"