Paste a passage,
see how “AI” it reads.

Where this tool comes from

The rules come from our official character-card copy review: a real user once judged a 91-point card “obviously AI” on sight and named the exact sentence-level tells — quote-as-punchline, antithesis flips, invented concept words. We turned those tells into quota-based rules and calibrated against human-written hit cards for zero false kills (human samples pass clean). Weighting: stock phrases score low, sentence molds score high — humans use isolated phrases too; it's the molds that fingerprint a machine.

What it can't catch (the honest part)

Templated AI copy with perfect structure, uniform detail density, and zero idle sentences passes these rules with a full score — the deepest layer of AI flavor lives in whole-text texture, beyond both rules and LLM judges. We ran a double-blind panel of four LLM judges on human-consensus anchors: 12% gold accuracy, with the judges unanimously calling human-written hits “AI.” So treat this tool as a lint pass, not a truth oracle: a low score means the text hits many patterns readers commonly flag as machine-flavored and deserves a line-by-line look; a high score guarantees nothing.

Full experiment data: our LLM judges called human writing AI