Does ChatGPT have a watermark?
For years the answer was no. In October 2026 that changed, but only in some places, and only for text you leave mostly unchanged. Here is where things stand.
Yes, in the EU. OpenAI is adding a hidden watermark called textGrain to ChatGPT and Codex text, starting in October 2026. Outside the EU it is not on by default, but API customers can turn it on. ChatGPT images also carry C2PA data. Hidden characters in ChatGPT text are not a watermark.
The answer, place by place
| Where you use it | Text watermark |
|---|---|
| ChatGPT in the EU | Yes, textGrain, starting October 2026 |
| Codex in the EU | Yes |
| ChatGPT outside the EU | Not by default as of 10 October 2026 |
| Apps built on the OpenAI API | Only if the developer turned it on |
OpenAI said textGrain is not on everywhere “at launch”. That could change later. We will update this page when it does.
What the watermark is
It is a pattern in the words ChatGPT picks, made with a secret key while the text is written. You cannot see it. It stays when you copy and paste, because it is part of the words. Read what textGrain is for the full story.
The hidden character myth
Before textGrain, people found odd characters like narrow no-break spaces (U+202F) or zero-width spaces (U+200B) in ChatGPT answers and called them a watermark. They are side effects of formatting. You can find and remove them with our text checker. They are not how OpenAI marks text.
If you are a student
- Your school cannot check for textGrain by itself. Only approved researchers and groups can use the detector.
- AI detectors that judge style make mistakes, also with people who write in a second language. A flag is not proof.
- Keep your notes, drafts and document history. They are the best proof of your own work.
- Follow your course’s AI rules. Whether you were allowed to use AI matters more than any watermark.
If you publish content
The watermark is not a label, and it does not stop you from publishing. Under the EU AI Act, you need a visible label for deepfakes and for AI text that informs the public on matters of public interest, unless a human editor checked it. Our AI Act checker takes two minutes.
If you build on the API
textGrain is off unless you turn it on for a supported model. If your users are in the EU, the AI Act’s marking rule may apply to you. Turning on textGrain may help with that for text.
Questions people ask
Does ChatGPT add hidden characters to mark text?
No. textGrain adds no characters. If you find zero-width spaces or narrow no-break spaces, they come from formatting or copying. You can remove them, but that does not change textGrain.
Will rewriting ChatGPT text remove the watermark?
OpenAI’s tests show heavy editing makes it much weaker. But only OpenAI can check if a text still has it, so there is no way to be sure.
Does a VPN change whether I get the watermark?
OpenAI has not said how it decides who is an EU user. Do not rely on guesses either way.
Do ChatGPT images have a watermark too?
Images made in ChatGPT carry C2PA content credentials. This is signed data inside the file that says it was made with OpenAI tools. It is separate from textGrain, and it can be lost when someone takes a screenshot or saves the image again.
Can AI detectors like Turnitin see textGrain?
Not unless they are approved to use OpenAI’s detector. Normal AI detectors judge writing style, which is a different and much less reliable method.