Free watermark check

ChatGPT watermark checker

Scan text from ChatGPT for invisible characters — and find out what is actually known about OpenAI watermarking.

Paste text

Paste, do not retype — invisible characters only survive a copy.

Upload PDF / DOCX / TXT

Click to choose a file

No account, no word limit

We never ask for a name, an email address or a university.

No invented number

Separate findings, not one percentage. Style is never counted as evidence.

Says what it cannot do

Claude's statistical watermark is checkable by nobody today, including us.

The short answer

As of August 2026 there is no deployed, checkable ChatGPT text watermark. OpenAI has signed the EU Code of Practice and stated it intends to extend provenance marking to text, but has not said how it will do so or when, and systems already on the market have until 2 December 2026 to comply. So ChatGPT text today generally carries no watermark to find — what it does carry is invisible Unicode picked up from the web interface, and this free checker finds that exactly, with no account needed. Em dashes are not a watermark; they are punctuation.

Free, no accountFinds hidden UnicodeNo invented verdictNo word limit

What this check establishes

  • Invisible Unicode characters copied along with ChatGPT output
  • Hidden messages encoded in zero-width characters
  • Mixed-alphabet words
  • How often em dashes, en dashes and curly quotes appear — reported as style, never scored

What it cannot establish

  • Whether text came from ChatGPT. No public OpenAI watermark exists to check.
  • That a text is human-written, from a clean result
  • Anything about authorship from punctuation habits

How to check ChatGPT text for hidden characters

  1. 1

    Paste your text, or drop in the file

    Copy rather than retype — invisible characters only survive a copy-paste. PDF, DOCX and TXT are read locally, and a DOCX also gives up its editing time and save count.

  2. 2

    No account, no word limit

    You are never asked for a name, an email address or a university. The text is analysed in your browser and the analysed text is then stored on our EU servers for twelve months, never linked to your identity.

  3. 3

    Read each finding separately

    You get one card per layer, each with its own confidence, and no combined percentage. Character findings are definite. Style habits are not evidence and are labelled as such.

Does ChatGPT watermark its text?

Not yet, in any deployed and checkable form. OpenAI has said for years that it built a statistical text watermarking method and held it back, giving two reasons worth taking seriously: paraphrasing defeats it, and detection errors fall hardest on people writing in a second language.

That position is now shifting. OpenAI is among the roughly 190 signatories to the EU Code of Practice and has said it intends to extend its provenance measures to text — but it has not stated the method, has not committed to Anthropic's approach, and has not shipped anything a third party can check. Systems already on the market have until 2 December 2026 under the Article 50 transition period, so this answer may change before the end of the year. It is checked here and dated for that reason.

Claude, by contrast, began marking output in August 2026. So 'is there a ChatGPT watermark' and 'is there a Claude watermark' currently have different answers.

Then why do people find hidden characters in ChatGPT output?

Because web interfaces put them there. Zero-width joiners, non-breaking spaces and narrow no-break spaces come from HTML rendering, from emoji sequences, and from the way text is laid out in a browser — and they survive a copy-paste into Word. Some of them also appear in text copied from any website at all, which is why this is worth checking before you submit a document regardless of where the words came from.

None of that identifies ChatGPT. It identifies a browser.

Em dashes are not evidence

The em dash became the internet's favourite AI tell, and it is a bad one. Microsoft Word converts double hyphens into em dashes automatically. Professional copy-editors use them. So do the style guides most universities hand out. Punctuation frequency tells you about a text's editing history, not its authorship.

This checker counts your dashes and quotes because people ask, shows them under a heading that says they are style rather than evidence, and refuses to fold them into a score. Turning habits like these into a confident percentage is precisely how AI detectors started accusing students who had written every word themselves.

Frequently asked questions

Not in any released, checkable form as of August 2026. OpenAI built a text watermarking method years ago and chose not to deploy it, citing how easily paraphrasing defeats it and the risk of unfairly flagging non-native English writers. It has since signed the EU Code of Practice and said it intends to extend provenance marking to text, without stating a method or a date; systems already on the market have until 2 December 2026 to comply with Article 50. So there is currently no ChatGPT watermark for any tool to detect, and this is a fact with an expiry date.
Almost always because you copied it out of a web page. Browsers and HTML rendering introduce zero-width joiners, non-breaking spaces and narrow no-break spaces, and these survive a paste into Word. The same happens copying from any website. It is worth cleaning up because these characters break LaTeX, citation tools and text search — but it does not indicate AI authorship.
No. Word inserts em dashes automatically from double hyphens, professional editors use them routinely, and most academic style guides permit them. Punctuation frequency reflects a document's editing history, not who wrote it. This tool reports dash and quote counts as style information and deliberately never scores them.
Yes, with no sign-up and no word limit. The analysis runs in your browser; the analysed text is then stored anonymously on our EU servers for twelve months and deleted automatically.

Sources

Related checks

Claude watermarkInvisible charactersFree watermark checkerAI check

The real question: will your work be flagged as AI?

That is what universities actually run, and those detectors are wrong about human writing more often than they admit. Our AI check publishes its false-positive rate beside every verdict. Free for the first 1,500 words.

Run the free AI check