AI detector, free and private
Paste a text to see if AI wrote it. The tool looks for leftovers from AI chats and hidden characters, then scores the writing. The text never leaves your computer.
The first check downloads about 125 MB, once.
How it works
- 1Paste the text
80 words or more. An essay, an email and a post all work. English gives the best results.
- 2Read what it holds
Text copied out of an AI chat often carries codes, tracking links and invisible characters. The tool lists every one it finds.
- 3Score the writing
A model trained on writing by people and by AI scores how much the text reads like AI.
How far to trust the result
Chat leftovers are the best evidence. Codes like turn0search0 only show up when someone copies an AI chat's answer.
The score is an educated guess. It caught a quarter of the AI texts in our test and flagged 2 of 990 texts by people. It misses most text that a person or a tool rewrote.
Never use a score alone to accuse someone.
For a better check
- Paste the whole text. Longer texts get steadier scores.
- If only part may be AI, check that part on its own. Give it 150 words or more.
- Compare it with the person's earlier writing. A sudden change in voice tells you more than a score.
- Ask about the work. Someone who wrote it can explain their choices.
Made by the team behind Cedric
Cedric is an AI teammate in Slack. He joins your Google Meet, Zoom and Teams calls, posts the minutes when the call ends and follows up on every to-do.
More free tools
See all tools →FAQ
Yes. There is no account and no daily limit. Your own device runs the check, so a text costs us nothing.
No. The text stays in this browser tab. The page downloads the checker once, about 125 MB, and never sends your text anywhere.
We tested it in October 2026 on 2,172 English texts. 1,182 came from 68 current AI models, some rewritten by humanizer tools, and 990 were written by people. At a score of 90 or more it caught 24% of the AI texts and wrongly flagged 2 texts by people. From 50 it caught 34% and flagged 31.
Text copied with ChatGPT's leftovers in it, yes, every time. Without them it comes down to the score. In our test it gave 40% of texts from current ChatGPT models and 44% from Gemini a score of 50 or more.
Yes, both ways. In a 2023 Stanford study, seven detectors flagged 61% of essays on average by non-native English speakers as AI. We set this tool's bar high to avoid that. In our test it gave 9 of 266 texts by English learners and non-native authors a score of 50 or more, and none reached 90.
Rarely. It gave 27% of rewritten AI text a score of 50 or more. It does spot lookalike letters, a trick some humanizer tools use.
Characters you can't see, like zero-width spaces and narrow spaces. Some AI models write narrow spaces, and some apps add zero-width ones when you copy. Alone they prove little. The tool lists them and can give you a copy without them.
Google hides a watermark in text from Gemini, and other AI companies have announced their own. Only the company that adds a mark can read it, and none has a public checker for text yet.
English. It reads other languages, but the score means much less.
Your AI employee is waiting for you
Get started for free100K credits ($50) on sign up. No card needed.