unmark

Remove the Claude watermark from text

Anthropic has said future Claude models will watermark their output with a SynthID-style statistical method — on by default, with no opt-out, to help meet the EU AI Act. Unmark rewrites Claude's text so that watermark's pattern is scrambled while your meaning stays intact, and strips any invisible characters in the same pass.

Paste AI text0 chars
Watermark-free text0 chars
Your watermark-free text appears here.
Strength
Humanise — natural rewrite, same length & meaning.
Paste AI text, then remove its watermark.
▸Hidden-character optionscore invisibles, tags & private-use are always removed

How Claude's watermark works

It's a statistical, token-level watermark: it subtly biases which words Claude picks so the output follows a measurable pattern, without any visible change to the text. Anthropic notes that light editing may leave the mark intact, but a complete rewrite removes it — which is exactly what Unmark does.

How to remove it

Paste your Claude text below and run a rewrite. Unmark rewords the text enough to disrupt the statistical pattern while preserving your meaning, then re-scrubs the result for invisible characters. Copy the clean output.

FAQ

Does Claude watermark its text?+
Anthropic has announced that future Claude models will watermark their text with a SynthID-style statistical method, applied by default with no opt-out to help comply with the EU AI Act.
Can I turn Claude's watermark off?+
No opt-out has been announced. The practical way to remove it from a given piece of text is to rewrite that text, which is what Unmark does.
Is the result guaranteed undetectable?+
No. Rewriting strongly degrades a statistical watermark, but no tool can promise text is permanently undetectable. Treat it as best-effort and re-read the result.

Related guides