Analyzer · AI Detection
The AI detector that shows its work
Every flag comes with the measured signals behind it — and the human counter-signals pointing the other way. A per-passage heatmap instead of a single scary number, and a vocabulary that affirms human writing instead of grading it on an accusation ladder.
Per-passage heatmap
Documents are scored passage by passage, not as one blob. Flagged passages show exactly which measured patterns fired; quoted material and short lines are excluded rather than mis-scored.
Two-sided vocabulary
Passages read from READS HUMAN through INCONCLUSIVE to STRONG AI-TYPICAL. The middle band is honestly gray — no verdict — and never counts against your document.
Signals, never accusations
Verdicts name patterns ("AI-typical"), never authorship. Formal registers — legal clauses, academic citations, statutes, templates — get structural protection because formality is not evidence.
No verdict, no charge
If the engine concludes there isn't enough scoreable prose for an honest verdict, the scan is free. Uploads are always free — the only step that costs credits is the Detect button, and it shows the price before you click.
Suggest edits on flagged passages
One click rewrites only the flagged passages — in your own voice, anchored to the passages that read human — with every changed word shown as a reviewable inline edit.
How it works
01
Paste or upload
Up to 30,000 words — DOCX, PDF, or plain text. Uploading is free and the document lands in your library as an original.
02
Detect
The engine blends a language-model read with measured stylometrics. Neither can convict alone — flags require corroboration.
03
Read the why
Open any passage to see the measured signals, the human counter-signals, and the model's corroborating impressions — labeled as exactly that.
What it doesn't claim
It catches text that reads like an unedited AI draft (0.87 AUC, measured monthly). It does not reliably catch AI text prompted casually or run through a humanizer — our testing says nothing affordable does — so a clean scan reports "no AI-draft signals found", never "human-written". The false-positive budget on human writing is 1% by design.
Straight answers
Why did my formal document score in the gray band?
Formal registers — contracts, statutes, academic prose — share surface patterns with AI drafting. The engine deliberately reports these as INCONCLUSIVE (no verdict) rather than flagging them, because uniform rhythm is register, not evidence. Gray never counts against the document.
Can it prove my text is human?
No detector can, and we say so on every scan. A clean result means no AI-draft signals were found. Tools claiming to certify human authorship should show you their false-positive curve.
What happens to quoted material?
Passages that are substantially quotation — statutes, block quotes, cited definitions — are marked QUOTED and excluded from scoring. A quotation is a verbatim copy of someone else's words; authorship analysis doesn't apply.
The other instruments
Free to start — 50 credits a month, no card.
AI Detection is one of six instruments sharing one document — check, verify, and polish without pasting twice.