Does ChatGPT Watermark Your Text Now? What OpenAI's EU Watermark Means (October 2026)

TL;DR: Soon, if you're in the EU. On October 5, 2026, OpenAI said it will add an invisible watermark, called textGrain, to eligible ChatGPT and Codex text generated in the European Union "over the coming weeks." Outside the EU, ChatGPT text is not being marked; developers anywhere can opt in through the API, where it is off by default. The mark shows only that an OpenAI system generated or processed part of the text, and for now only approved researchers and expert organizations can run the detector.

---

For two months, Claude was the only major chatbot stamping an invisible mark on the text it writes. On October 5, OpenAI said ChatGPT is next, at least in Europe. If you use ChatGPT for drafts, emails, or reports, here is what changes, for whom, and what it does and doesn't mean for how you work.

What OpenAI announced

OpenAI published a post titled "Our approach to EU text provenance rules" and a technical report written with researchers from the University of Pennsylvania and Yale. The plan has two parts:

• ChatGPT and Codex in the EU. Over the coming weeks, OpenAI will add an invisible watermark to eligible text generated in the EU. TechCrunch reports this covers all ChatGPT and Codex plans there. OpenAI did not give a start date or say what makes text "eligible."
• The API, worldwide. From October 5, API customers anywhere can turn on text watermarking for select models. It is off by default.

The trigger is the EU AI Act. Its transparency rules, which took effect August 2, 2026, require AI providers to mark generated content in a machine-readable way. OpenAI, Anthropic, Google, Meta, and Microsoft have all committed to the EU's code of practice on marking AI-generated content.

Will my ChatGPT text be watermarked?

It depends on where the text is generated:

Where you use it	Watermarked?
ChatGPT or Codex, text generated in the EU	Yes, rolling out over the coming weeks
ChatGPT or Codex, text generated outside the EU	No, under this announcement
OpenAI API, anywhere	Only if the developer turns it on (off by default)

Coverage of the announcement does not describe a way for EU ChatGPT users to switch the watermark off. If you work for an EU organization, or travel there, assume ChatGPT drafts made there will carry it once the rollout reaches your account.

How the watermark works

You won't see anything. textGrain doesn't add hidden characters or a label. Instead, it changes how the model picks between words that would fit equally well. A secret key steers those small choices, and across a long enough passage, a detector holding the same key can pick up the pattern. OpenAI's technical report describes the math in detail and says the detector needs only the text and the key.

Because the signal lives in the words themselves, it survives copy and paste. That is the same basic approach Anthropic uses for Claude.

What the watermark can and can't tell anyone

OpenAI is unusually direct about the limits. A detected watermark can indicate that an OpenAI system generated or processed part of the text. It cannot tell:

• who used the system,
• how much a person contributed,
• who owns the text, or
• whether the content is accurate.

So a report you wrote yourself and asked ChatGPT to tighten could test positive, just like a report ChatGPT drafted from scratch. The detector can't tell those apart.

It also misses things. OpenAI says short passages and later edits make detection less reliable, and translated or math-heavy text is harder to catch, according to TechCrunch. The figures OpenAI published, as reported by TechCrunch and The Decoder, show detection falling steeply once a share of the words are swapped for synonyms. OpenAI says the absence of a watermark does not prove a human wrote something.

Who can check for it

For now, almost no one. OpenAI says access to the detector will initially be limited to approved researchers and expert organizations, and it cites the reliability limits above as part of the reason. There is no public "was this written by ChatGPT?" checker from OpenAI. A colleague, client, or professor can't test your document against the watermark today.

How this compares with Claude

Anthropic went further, sooner. Since August 2026, text from Claude models released on or after August 2 carries a watermark worldwide, across the Claude apps, Claude Code, Cowork, and the API, with no opt-out. OpenAI is marking ChatGPT and Codex only in the EU and leaving the API choice to developers. We covered Claude's version in Claude now marks all text it generates and what Claude's watermark can and can't detect.

What to do if you use ChatGPT at work

1. Know where your text is generated. If you or your team work in the EU, expect ChatGPT and Codex output to carry the mark once the rollout reaches you.
2. Keep disclosing the way you already do. The watermark doesn't replace your employer's AI policy, a client contract, or your professional rules, and it doesn't change them. See how to review an AI draft before you send it and the state-by-state AI disclosure rules.
3. Don't treat a detector result as a verdict. If someone flags your work, a positive result means AI touched part of it, not that AI wrote it. A negative result proves nothing either way.
4. If you build on the API, decide deliberately whether to turn watermarking on, and tell your users what you chose.

---

Sources

• OpenAI, Our approach to EU text provenance rules (October 5, 2026; announcement also posted on the OpenAI Developer Community)
• OpenAI, University of Pennsylvania and Yale, textGrain: Entropy-Calibrated Watermarking for Language Model Text (technical report, October 5, 2026)
• TechCrunch, OpenAI will start watermarking ChatGPT's text in the EU (October 5, 2026)
• The Decoder, OpenAI will watermark ChatGPT text in the EU but makes it optional for API users worldwide (October 5, 2026)