Does Claude watermark its text?
Yes. Anthropic began marking Claude's text output in August 2026 and applies it globally, across the Claude apps, the Claude Platform API, Claude Code, Cowork and Claude Tag, and on the cloud platforms that resell Claude.
The company gives its reason plainly: the EU AI Act requires machine-readable marking of AI output, and Anthropic chose one worldwide rollout because, in its words, "we don't yet have a durable way to scope it by region."
The help center's model table lists text watermarks for every current model, from Haiku 4.5 up to Opus 5.5 and Fable 5.1. The API release note of 1 September adds that marking "requires no changes to your requests or response handling," which means developers see nothing different in the output.
Sources: Anthropic, Claude text watermark explainer · How Claude marks AI-generated content · API release notes, 1 Sep 2026
Why can't a character cleaner remove the Claude watermark?
Because there is no character to clean. Anthropic states that nothing is inserted into the text: the watermark only steers which of several equally good next words Claude picks, using a secret key.
"Nothing is added to the text and there are no hidden characters."Anthropic, Claude text watermark explainer
The method is, per Anthropic, "a version of the SynthID-Text approach" published by Google DeepMind in Nature in 2024. Change enough of those word choices and the statistical pattern fades; leave them and it stays, however many zero-width spaces you delete.
Stray invisible characters can still turn up in text that passed through other apps, so the second tab strips them for free. It just is not the Claude watermark.
Sources: Anthropic explainer · Dathathri et al., Nature 2024 (SynthID-Text)
What actually removes the Claude watermark?
A full rewrite. Anthropic's own answer is that light editing "probably won't remove the watermark completely; a complete rewrite where every word is replaced will."
Two more lines from Anthropic help explain how much is left after a given change. Short passages carry "fewer word choices and thus less information," and factual text, proofreading and code carry a sparser mark. A translation made by Claude is still marked, because Claude chose every word of it.
The panel below shows the difference on a sample paragraph. Yellow marks every run of three words that survived from the original. That is the raw material a word-choice detector reads.
What does a detected Claude watermark prove?
Only that Claude was likely involved at some point. Anthropic says the mark "cannot distinguish 'Claude wrote this' from 'Claude heavily edited this'," and it carries nothing about the person who used it.
The company is equally direct about the other way round: a missing mark "doesn't mean the content wasn't AI-generated or processed." People use Claude to proofread, translate and summarize, so a positive result can come from light help, too.
On ownership, Anthropic writes that the watermark "doesn't say anything about ownership or authorship, and doesn't change a user's rights under our terms." Rules from a school, employer or publisher about AI use and disclosure are a separate question and still apply.
Sources: Anthropic explainer · Claude help center
Who can check text for the Claude watermark?
A short list of organizations, through a private-preview detection API: regulators, law enforcement, media and fact-checkers, independent researchers, educational organizations, EU civil society groups, and companies that must verify marking for their own compliance.
There is no public text checker. The free Claude Content Checker reads the signed metadata Claude attaches to files such as images, not text. So a website offering to "detect the Claude watermark" in pasted text cannot be checking the real mark, and neither can we. What we can measure is how much of your original wording a rewrite left behind, and the receipt shows that number.
What does the Claude watermark remover cost?
Stripping characters costs nothing. Sign in with your email and the first four rewrites are free; paid plans run from $12 a month on yearly billing.
Basic
- 80 rewrites every month
- 1,000 words a rewrite
- Receipt with each one
$18 charged monthly until you cancel.
Pro
- 200 rewrites every month
- 150,000 words across them
- 1,200 words a rewrite
$27 charged monthly until you cancel.
Ultra
- 250 rewrites every month
- 300,000 words across them
- 3,000 words a rewrite
$48 charged monthly until you cancel.
USD. Stripe processes payment as seller of record and adds tax where it applies. Cancel from your account at any time.
Who runs this Claude watermark remover?
A small, independent computational linguistics studio. Rewrites run on an engine that learned how people write from more than 2,500 essays by real students, rather than from text other AI models produced.
Stripping happens on your device. A rewrite passes your text to our server and the engine, and the result is deleted after 24 hours. Quotes on this page are copied from Anthropic's own pages, linked where they appear.
Independent site. Not affiliated with or endorsed by Anthropic. Claude is a trademark of Anthropic. Contact: support@claudewatermarkremover.pro