What this site does
You paste text or upload a TXT, DOCX or PDF file, and the checker returns a verdict, the share of the text flagged as AI-like, a confidence level and sentence-level highlights. It works on text only. Detection is performed by the Pangram Labs API, and we build the experience around it: clear inputs, a readable result and plain explanations.
Why we built it
People now ask whether a piece of writing came from a person or a model: a teacher reading a stack of essays, an editor reviewing a submission, a writer checking their own draft. Many tools answer with a single score and little context. That invites over-confidence, and over-confidence can hurt people, especially when a human text is flagged by mistake. We wanted a checker that shows where the signal comes from and says openly how much weight it can bear.
What we believe about results
- A detector result is a signal, not proof.
- Nobody should be accused or penalised on a detector result alone.
- Limits should be stated next to the tool, not buried in a footnote.
- If we have not measured something, we do not claim it. That is why you will not find an accuracy percentage here.
Who it is for
Teachers and students who want to review writing with context, writers and editors who want to see how a draft reads, and anyone curious about a piece of text. If you use it to assess someone else's work, please read the guidance on the accuracy page first.
How to learn more
The how it works page explains detection in plain language, the methodology page covers what we analyse and where results can mislead, and the privacy page follows your text from submission to result. The rules for using the checker are in our terms.