AI content detector, honest by design
See the writing-style signals in your text as plain bands, never a made-up "82% AI" score. AI detection is unreliable, and we show you exactly why instead of pretending otherwise.
Analysis runs locally in one pass, and your text is not stored. This tool does no web search; for copied text use the plagiarism checker.
Read this first: AI-writing detection is not reliable, and this tool does not pretend otherwise. It reports stylometric signals (patterns common in generated text), not proof. False positives are common: careful, formal, edited, or non-native writing often trips these signals, and a determined author can evade them. Never use this as sole grounds for an accusation.
What the signals mean
Stylometry, not magic
We measure well-understood traits (em-dash habits, "not just X, but Y" phrasing, hedging, chat-assistant leftovers, sentence-length variation, model-favored vocabulary) that tend to differ between human and generated prose.
Bands, never fake percentages
There is no honest way to turn writing style into a calibrated probability. So we show a coarse band and the individual signals, and let you judge, instead of an invented number.
False positives are common
Formal, edited, technical, or non-native writing often looks "AI-like" to any detector. Never treat a band as proof or grounds for an accusation.
A clean result is not proof of anything either
Newer models write less like the early-2020s stereotype, and light editing removes most of what any style test can see: changing a few dashes and adding contractions is often enough. So this tool can miss generated text, and "no strong AI signals" does not mean a person wrote it. We would rather tell you that than sell you a number.