Does ChatGPT Watermark Its Text?

Do not treat every ChatGPT response as watermarked. A watermarking technique, a product rollout and an account's current behavior are separate facts.

Official announcement, access guide and technical report reviewed October 6, 2026. Availability below distinguishes announced rollout from verified access. This site is independent and does not offer a TextGrain detector.

Check product and region separately

Product or context	What to verify	Official status checked October 6, 2026
ChatGPT in the EU	Eligible ChatGPT and Codex output: phased EU rollout over the weeks following the October 5 announcement.	Announced rollout; not evidence every account is enabled.
ChatGPT outside the EU	The announcement does not make text watermarking a global ChatGPT default at launch.	No worldwide-default claim.
OpenAI API opt-in	Worldwide API customers can opt in for select models. Watermarking remains off by default.	API opt-in announced October 5; verify your model and settings.
Older responses	Generation date and settings. Publication of a method does not retroactively alter existing text.	No historical blanket claim

Does copy and paste remove it?

Copying unchanged words does not undo a statistical pattern in token choices. Clipboard software may change formatting or Unicode characters, but those changes are not proof about a statistical watermark. The OpenAI technical report describes a signal coupled to token generation, not a removable file label.

What about edits and short excerpts?

Editing changes the observed token sequence and the contexts used by a keyed test. Short excerpts give the detector fewer positions to score. Neither editing nor a small sample provides a guaranteed removal method or a reliable human-authorship conclusion.

Can a general AI detector confirm it?

No. A classifier's AI score is not a TextGrain watermark result. A real watermark test requires the matching setup. Read the detector access and limitations guide before interpreting either type of result.

What can I check myself?

You can inspect hidden characters locally, review the technical mechanism, and consult the source links for current rollout information. This site offers no score, official account status lookup or claim that your particular response contains a watermark.