AI-written (Python, Claude)
Reads as AI-generated
Both channels: clearly AI (90 / 95)
Free AI code detector
Paste a single file and get a reading from two independent channels, with the known limits stated up front. No sign-up, no storage of your submission beyond your own history.
Free checks: 3 per day. Code detection is newer and less certain than prose detection — we show you what the channels said, not a made-up certainty.
Free scan: we analyze your first ~300 characters
We keep your document so you can find it again, and you can delete it whenever you want. We do not sell it. Manage or delete
Three labeled samples from our own probe runs (August 2026, frozen inputs, real API calls). The readings below are what the channels actually returned.
AI-written (Python, Claude)
Reads as AI-generated
Both channels: clearly AI (90 / 95)
Human-written (C, git v1.7.0, 2010)
Reads as human-written
Both channels: clearly human (0 / 5)
AI-written (JavaScript, Claude)
Uncertain — and we say so
Channels disagreed (60 / 95)
Disagreement between channels is shown as disagreement. We do not average two conflicting readings into a confident-sounding number.
Measured
In our frozen retest (49 real code files, three AI generators), the channels behind this page separated AI-written from human-written code cleanly on that set. Mixed or lightly edited files were harder — that is in the data too.
Our estimate
Short snippets and heavily templated code (boilerplate, CRUD, config) read as more machine-like regardless of who wrote them. Treat borderline readings accordingly.
What no one can know
No detector can prove who wrote a file. If you were flagged by a tool somewhere else, the persuasive evidence is your process: drafts, history, and being able to explain the code — not a cleaner score.
We do not rewrite code to beat a detector, and we do not sell a score that claims to prove authorship. If your code was flagged and you wrote it, we would rather help you document how you wrote it than sell you a number.