ChatGPT Watermark Detector
Paste an answer from ChatGPT to see whether it carries hidden characters: zero-width spaces, narrow no-break spaces, direction marks and other invisible Unicode that can act as a watermark. The detector reports exactly what it finds, including when it finds nothing.
What a ChatGPT watermark can look like
There are two kinds of text watermark. A character watermark hides invisible Unicode characters in the text. A statistical watermark adds no characters at all: it shifts which words the model picks, and a detector looks for that pattern.
- Statistical: OpenAI has announced textGrain, an invisible statistical signal in the model’s word choices. It says it is adding it to eligible ChatGPT and Codex text in the European Union, and that API customers can opt in for select models. Access to its detector is limited to approved researchers and expert organizations. No character scanner can see this watermark, this one included.
- Characters: some users have reported ChatGPT answers containing narrow no-break spaces (U+202F) and other special spaces. Whatever put them there, they are real characters in your text, and this detector flags them.
How to check ChatGPT text
Copy the answer with ChatGPT’s copy button or by selecting it, paste it above, and read the verdict. You will see the count and name of each hidden character. If any are found, copy the cleaned version, or use the rewrite option to rephrase the text.
What this detector does not do
It cannot read OpenAI’s textGrain watermark, and it does not decide whether a human or ChatGPT wrote the text. Hidden characters show what is in the text, not who wrote it: they also come from web pages, documents and other apps. The AI-likelihood score lists its signals openly, and its style signals are only indicative.
Frequently asked questions
Does ChatGPT add a watermark to its text?
OpenAI says it is adding textGrain, an invisible statistical watermark in word choice, to eligible ChatGPT text in the European Union. That watermark adds no hidden characters, so this tool cannot detect it. What this tool does detect is any hidden or special Unicode characters in the text you paste.
Can this detect if text was written by ChatGPT?
No tool can say that for certain. This one reports concrete hidden characters and a transparent AI-likelihood score. Text with no hidden characters may still be AI-written, and text with them may not be.
Does it work for every ChatGPT model?
Yes. It checks the characters in the text itself, so it works the same for any ChatGPT model and for Claude, Gemini or any other tool.