Free AI-text estimate
Was this text likely written by AI?
Paste at least 50 words of English. A statistical model gives an estimate - not proof - and shows the words that pushed it either way. It runs in your browser: your text is never sent to us.
Pulls toward AI
Pulls toward human
What this estimate can and cannot tell you
| Tested on | AI caught | Human wrongly flagged |
|---|---|---|
| Texts like its training data | 99% | 1.7% |
| A topic it never saw (finance) | 97.5% | 2.2% |
| Fiction | 6.6% | 0% |
| Text from an AI model family it never saw | 39% | 3% |
| Independent benchmark: 11 AI models, 8 kinds of text | 33% | 3% |
| Research abstracts in that benchmark | 37% | 7% |
So: when it says AI on familiar kinds of text, it is usually right; when it says human, that proves little, because it misses most text from newer or unfamiliar AI models. On research abstracts it wrongly flags about 7 in 100 human-written ones.
Never use this score alone to accuse a student or an author. It is untested on non-English text, on heavily edited AI text and on the newest AI models.
A free tool, separate from our rewriting service. Trained on the public HC3 (Guo et al., CC BY-SA 4.0) and Ghostbuster (Verma et al., 2023) datasets.
How we think about academic integrity