Understand the signal
ChatGPT watermark: how it works
A ChatGPT text watermark is a statistical pattern in generated wording. It is not a visible label or a record of who wrote the text.
What is a ChatGPT text watermark?
OpenAI calls its method textGrain. It links token choices to keyed randomness during generation; a matching detector tests for the resulting statistical dependence. A token can be a word or part of one. This is a provenance mechanism built into generation, rather than a judgement based only on whether the writing sounds artificial.
Sources: textGrain technical report (PDF) ↗
What did OpenAI announce?
On 5 October 2026, OpenAI announced watermarking for eligible ChatGPT and Codex text in the EU over the following weeks. API customers worldwide can opt in for selected models; API watermarking stays off by default. These scopes should not be read as a watermark on every ChatGPT answer everywhere.
- ChatGPT and Codex: an announced EU rollout for eligible output.
- API: opt-in, selected models, customers worldwide.
- Detector: separate access requirements, not an automatic public release.
Sources: OpenAI rollout announcement ↗
What could a result actually tell you?
An OpenAI signal is a narrow provenance finding. It does not identify an author or account, measure the human contribution, validate the content, or decide whether AI use was allowed. Conversely, an absent signal is not a certificate of human authorship. Keep the detector result separate from any judgement about a person.
Sources: OpenAI rollout announcement ↗
Why context still matters
Two practical questions belong alongside any future check: which version of the text was checked, and which question are you trying to answer? For an academic draft, keep your notes, sources and version history. These can explain the writing process more directly than a single automated label. Follow the assignment’s own disclosure requirements when describing AI assistance.
Common questions
Can I see a ChatGPT watermark in my document?
No visible stamp is added by textGrain. The signal sits in word-choice patterns, rather than in hidden characters or a document logo.
Does this apply to older ChatGPT text?
The announcement is not a retroactive marking of existing documents. A text’s creation date, product and watermark settings matter.
Can I check a watermark with SafeRender now?
No. SafeRender currently offers a waitlist, not a public text watermark check. Any launch depends on approved access and permitted use.
Sources and editorial notes
This is an independent explanation, based on the linked sources and their review dates below. OpenAI’s announcement describes a staged rollout; product documentation may describe the intended scope more broadly. We do not assume that every eligible output is already covered.
- OpenAI rollout announcement ↗OpenAI · Checked
- OpenAI provenance guide ↗OpenAI Help Center · Checked
- textGrain technical report (PDF) ↗OpenAI and research collaborators · Checked
Want to know when you can check your text?
Join the waitlist for a launch notification if SafeRender can offer the detector. No draft needed and no launch date confirmed.