ChatGPT Watermarks: OpenAI Announces textGrain for EU Rollout
Updated October 6, 2026. OpenAI has announced textGrain, a text watermarking system. Its October 5 announcement describes an EU rollout for eligible ChatGPT and Codex output over the coming weeks, alongside worldwide API access that customers can enable for select models. API watermarking remains off by default. It is not a global ChatGPT default at launch. Read OpenAI’s announcement.
For people who use ChatGPT to draft newsletters, client documents, or coursework, the useful question is what changes in their writing workflow. At WriteBros, we help people rewrite AI-assisted text into more natural prose and apply their own writing style. That editing role is different from certifying that a particular watermark has disappeared.
What OpenAI announced
| Where you use OpenAI | Announced text watermark coverage |
|---|---|
| ChatGPT and Codex in the EU | Eligible output, across plans, rolling out over the coming weeks. |
| OpenAI API | Opt-in for select models worldwide; off by default. |
| ChatGPT outside the EU | The announcement does not introduce a global default. |
For an existing document, keep your source draft and the date you created it. That record makes later comparisons easier and avoids relying on memory about which version you edited.
A ChatGPT watermark is not hidden formatting
OpenAI’s Help Center says textGrain works through statistical word choices, not hidden characters, invisible spaces, or unusual punctuation. The signal is in the wording. Copying the same text into a plain-text editor therefore does not address that mechanism. See OpenAI’s provenance explanation.
The textGrain technical report describes generation tied to a secret key and statistical detection using that key. A generic “AI probability” score is a different kind of evidence. If a tool reports that a paragraph sounds human, that alone does not show whether OpenAI’s watermark detector would recognize it.
Can rewriting remove a ChatGPT watermark?
OpenAI reports that editing can weaken detection. That is evidence about the watermark’s limitations, not a tested success rate for any commercial humanizer. We have not yet verified WriteBros against OpenAI’s production textGrain detector.
WriteBros can help with the part writers can inspect directly: revising the language. You can judge whether a rewritten paragraph explains your point, sounds appropriate for its audience, and keeps the original facts. Our ChatGPT watermark removal and rewriting guide explains how to approach that task without confusing a rewrite with a watermark test.
What creators, professionals, and students can do now
- Creators: Start with the idea you want readers to remember. Add a specific observation, remove a generic introduction, and revise the rhythm to fit your newsletter or post. Save the original draft so you can compare changes.
- Professionals: Check claims, dates, prices, and commitments before changing the tone. A smoother client email is useful only if it still promises what you intended. Review the final wording yourself.
- Students: Keep your argument, reading notes, and citations visible during revision. Use editing support within the rules for your assignment, and make sure you can explain the reasoning in the finished work.
For all three groups, an editorial review answers a more useful first question than a score: “Is this a clear, accurate draft that I am comfortable putting my name on?” Watermark detection is a separate assessment.
Where WriteBros fits
Use the WriteBros AI humanizer to revise a ChatGPT draft, then compare the output with your source paragraph. WriteBros also supports rewriting guided by your writing samples through its personal writing-style feature. That feature concerns written voice, not audio cloning.
Choose tools around the actual task: clearer wording, a more consistent voice, or an evidence-backed watermark assessment. Our AI humanizer buyer guide helps separate those selection criteria. To start editing, create a WriteBros account, or open your dashboard if you already have one.
Is the textGrain detector publicly available?
OpenAI’s Content Provenance documentation says text verification is available only to approved organizations. Its public image and audio verification tools should not be presented as a publicly available ChatGPT text detector.
We will update this guide when there is reproducible, relevant evidence for WriteBros and textGrain. Until then, the product’s rewriting capability and a verified watermark-removal result remain separate claims.