RealWriteAI
New

ChatGPT Watermark Remover

OpenAI is adding an invisible watermark, called textGrain, to ChatGPT and Codex text in the EU. It is a pattern in the words the model picks, so cleaning up characters does nothing to it. Paste your ChatGPT draft below and the humanizer rewrites it sentence by sentence in new words, keeping what it says.

Try it here without an account (50 to 100 words). Free accounts get 300 words a day.

Try

0 / 100 words

ChatGPT’s watermark at a glance

Who marks it
OpenAI, with a watermark called textGrain
Since
Announced 5 October 2026, rolling out over the following weeks
Where
ChatGPT and Codex in the EU. On the OpenAI API only if a developer opts in.
How
A secret key nudges which word comes next; a detector adds up hundreds of those nudges
Identifies you?
No, according to OpenAI
Who can check
Approved researchers and expert organisations only
What weakens it
Editing, translation and short passages, by OpenAI’s own account

Sources: OpenAI’s announcement as reported by TechCrunch and BleepingComputer, 5 October 2026.

Detection after editing, in OpenAI’s tests

Detection rate, 400-token passages

  • Unedited92%
  • 10% of words swapped66%
  • 25% of words swapped17%
OpenAI’s own figures at a 1% false-positive rate, as reported by BleepingComputer. The edits were synonym swaps, not full rewrites.

How it works

  1. 1

    Paste your ChatGPT text

    Paste a draft from ChatGPT, Codex or the OpenAI API. Without an account you can try 50 to 100 words; sign up free for longer drafts, file upload and voice input.

  2. 2

    Press Humanize

    The humanizer rewrites the draft with different word choices and sentence shapes. There are no settings to tune first.

  3. 3

    Check what changed

    A meaning check compares the rewrite with your draft and flags any number, name, quote or citation that moved. Click a sentence for two or three alternatives, or keep your original.

What OpenAI’s own tests say about editing

textGrain nudges the model’s word choices with a secret key, and a detector adds up hundreds of those nudges. OpenAI published how much editing it takes to wash the pattern out.

10% of words swapped

Detection: about 92% → 66%

Replacing one word in ten with a synonym already cut detection on 400-token passages by about a quarter, in OpenAI’s own test.

25% of words swapped

Detection: about 92% → 17%

At a quarter of the words, the detector found the mark in fewer than one passage in five. Short passages, math answers and translated text were already harder to detect.

Hidden characters

The mark is not made of invisible spaces or odd punctuation, so stripping them changes nothing. The humanizer rewrites every sentence instead of swapping a word here and there, then shows you what changed.

Figures: OpenAI’s tests on 400-token passages, as reported by BleepingComputer and TechCrunch, 5 October 2026.

Why RealWrite

We would rather tell you what the tool won’t do than oversell it.

What it does

  • Rewrites ChatGPT text in new words, sentence by sentence
  • Changes sentence length, rhythm and stock phrasing
  • Checks the meaning against your draft after every run
  • Flags every changed number, name, quote, citation, link or claim
  • Works on text from any ChatGPT model, Codex or the API

What it won’t do

  • Promise that the watermark is gone: OpenAI’s detector is not public
  • Remove anything from images or files; it works on text only
  • Just strip hidden characters and call it done
  • Change your school’s, employer’s or publisher’s rules on AI use

Who it’s for

  • Writers in the EU

    Once the rollout reaches you, longer ChatGPT replies carry textGrain. Rewrite the drafts you publish so they read like you and still say what you meant.

  • Students

    ChatGPT helped you outline or draft, and now you want the final text in your own voice. Check your course’s rules on AI use first; a rewrite does not change what they allow.

  • Codex users

    Docs, comments and release notes drafted with Codex in the EU are covered too. Rewrite the parts people read before they ship.

  • API developers

    If your app opted in to watermarking, the text it generates carries the mark. Rewrite what goes out under your brand in your house style.

Questions about the ChatGPT Watermark Remover

For the full background, read our guide: Does ChatGPT Watermark Its Text? OpenAI’s textGrain Explained

In the EU, yes, or soon. On 5 October 2026 OpenAI said it would add an invisible watermark to eligible ChatGPT and Codex text in the European Union over the following weeks, to meet the EU AI Act’s transparency rules.

Developers anywhere can turn watermarking on for supported models, but it is off by default.

OpenAI calls it textGrain. A secret key nudges which word the model picks next, and a detector that has the key adds up hundreds of those small nudges. Nothing visible is added, and the mark stays with the text when you copy and paste it. OpenAI says it does not identify the user.

Not yourself. OpenAI’s detector is open only to approved researchers and expert organisations for now. Sites that claim to detect it are guessing.

Because the mark is not made of characters. It lives in which words were chosen. Stripping spaces or punctuation leaves every one of those words in place.

We can’t promise either way, because nobody outside OpenAI’s approved testers can run the detector. What OpenAI’s own tests show: swapping 25% of the words for synonyms dropped detection from about 92% to 17%. The humanizer rewrites every sentence, wording and structure, instead of swapping a word here and there.

You can try it on this page without an account (50 to 100 words). A free account gets 300 words a day; paid plans raise the limit: Essential 30,000 words/mo, Pro 100,000 words/mo and Business Unlimited words/mo.

Rewriting a draft you were given is ordinary editing. Your school, employer or publisher may still have rules about AI help and about disclosing it, and those apply to the rewritten text too, so check them before you hand anything in.

More writing tools

See all tools

Compare word allowances on the pricing page, or read our writing guides on the blog for more on how AI text gets spotted and how to edit it.

Try the ChatGPT Watermark Remover on your own text.

Try it freeSee pricing