Big Update For ChatGPT Users: Your AI-Generated Text Could Soon Carry A Hidden Watermark; Here's Why

OpenAI is going to introduce an invisible watermark for ChatGPT and Codex text generated in the European Union, which will now be easier to identify whether content was created by artificial intelligence (AI).

The EU AI Act has now introduced new transparency requirements for AI-generated content. OpenAI announced the change yesterday, October 5, 2026, saying the new system will be rolled out to eligible ChatGPT and Codex users in the EU over the coming weeks.

What Is The ChatGPT Invisible Watermark?

OpenAI calls its technology textGrain. It will be different from the visible watermark as users will not see any special symbols, characters or markings in ChatGPT-generated text.

Instead, the system creates a statistical pattern in the way words are selected. This pattern can later be detected by a specialised AI text detector.

OpenAI said the watermark does not use hidden characters, invisible spaces or unusual punctuation.

The company also said the watermark should not have a meaningful impact on ChatGPT's performance or the quality of its answers.

Will All ChatGPT Users Get The Watermark?

No. For now, the ChatGPT text watermark will be introduced for users in the European Union. OpenAI has not announced the same default rollout for regular users outside the EU. However, OpenAI said API customers globally can opt in to text watermarking for select models. The company is also working on a detector that can identify whether OpenAI's watermark is present.

How Accurate Is The AI Watermark?

The watermark is not a perfect AI content detector. OpenAI said that, at a 1% false-positive rate, its testing showed around 80% detection for 200-token passages and about 95% detection for 400-token passages in psychology-related text. Shorter content, mathematical text, translations and heavily edited text can be more difficult to detect.

It also cannot determine how much of the final content was written by a human. It should also be noted that if the detector does not find a watermark, that does not automatically mean the content was written by a human.