Paste your text and get seven readings at once — how machine-like it is, where it overlaps with the web, how it reads, and what an editor would flag.
The detector reads four things that separate human drafts from machine drafts. Burstiness is the variation in sentence length: people write a long sentence, then a short one, while language models tend toward an even rhythm. Lexical diversity measures how often you reach for a new word instead of repeating one. Phrasing patterns looks for connectives models over-use. Opening repetition checks how many sentences start the same way.
Each sentence gets its own score, which is why the result is a highlighted document rather than one number. A paragraph that reads as machine-written is far more informative than a percentage attached to a whole essay.
A score of 90% does not mean 90% of your text was machine-written. It means the pattern resembles patterns the checker associates with machine drafts, at that confidence. Those are different claims, and conflating them is how people end up wrongly accused.
No AI detector is accurate enough to prove authorship, and any tool claiming otherwise is overselling. This one reads sentence rhythm, vocabulary spread and phrasing, then reports a confidence figure alongside the sentences that drove it.
It can. False positives cluster around non-native English writers, technical writing, and drafts edited several times. That is why every sentence carries its own score instead of one blunt verdict.
You can check up to 150,000 characters per scan and 75 files at once in batch mode. No daily cap, no account.
No. The analysis runs in your browser. Nothing is uploaded or added to a training set.