AI Detector

Because detection is probabilistic, not proof. Detectors look for patterns — even sentence rhythm, predictable phrasing, low variation — and human writing has those patterns too, especially clear, polished, or formal prose. A flag means your text resembles the patterns a detector associates with AI, not that a machine wrote it. That's exactly why we never show a naked score: every result pairs the likelihood with the uncertainty so you can see what stood out, revise it in your own voice, and re-check.

They compare your text against the statistical patterns typical of AI-generated writing — how predictable each word is, how uniform your sentence lengths are, how much variation there is across the passage. From that they estimate a likelihood. They don't read meaning or know who typed it, which is why the honest output is a likelihood with its uncertainty, never a verdict. We show you which passages stood out and why, so the estimate is something you can actually act on.

They're unreliable — that's why we never show a naked score, always show uncertainty, and publish our limits. No tool can promise certainty, and the ones that do are selling it. Human writing gets flagged, and different detectors disagree with each other. We'd rather be honest than scary: use the check to find what stood out and strengthen it, not to chase a number.

Because each is trained on different data and weighs the patterns differently, so they disagree — sometimes sharply — on the same passage. No detector can promise what another one will say. A check here tells you what our model noticed and how strongly, with the uncertainty shown; it can't predict another tool's verdict. What travels between tools is your own clear writing and the documented record that you checked it.

With evidence, not a score. Run your own check before you submit, see what stood out, revise anything unclear in your own voice, re-check, and keep your verification report as a dated record that you did the work and checked it. If a dispute happens, that record — not a promise from any single detector — is what changes the conversation.

No one can promise another tool's verdict — detectors disagree, and ours can't predict theirs. But a documented pre-check catches the passages you can strengthen and changes the conversation if a dispute happens. That's what the report is for.

Your documents are never indexed, resold, or used to train detection models.

No. The in-page check above needs no login — paste your text and see how likely it is to be flagged, with the uncertainty shown. You'll create an account only when you want to see every finding, the fixes, and your full verification report, or to run the Full Check on the whole document.