Claude Watermark Remover
Clean hidden Unicode marks first, then use a structural rewrite for statistical watermark patterns. Free to try.
How Claude marks AI-generated content
Anthropic documents this publicly, and it is worth reading before you trust any tool that claims to undo it — including this one.
A watermark in the text
Imperceptible to a reader, and Anthropic states it survives copying and some editing. It covers models released from 2 August 2026 across the Claude apps, the API, Claude Code and its cloud partners.
The cleanup on this page does not touch it. A mark that survives copying is not sitting in a character you can delete — it is in the writing, which is why rewriting is the only stage that could affect it, and why nobody can promise you that it did.
C2PA provenance metadata
For image files — .png, .jpg and .svg — Claude attaches digitally signed metadata following the C2PA open standard, the same provenance scheme used across the wider industry.
That one lives in the file rather than the words, so it never reaches a text box. This tool takes text only and cannot see it, let alone change it.
Anthropic also sets out the limits in both directions: a detected mark means content may have been processed by Claude rather than proving it, and an absent one proves nothing at all. Read Anthropic's own explanation.
Frequently asked questions
- What does the Claude Watermark Remover remove?
- It starts by stripping hidden Unicode characters such as zero-width spaces, word joiners, text-direction controls, tag characters and unusual spaces. Those can be removed exactly without changing the visible wording. If you also choose Rewrite, the tool changes sentence structure and word choice to address patterns that are carried by the text itself.
- Does Claude put invisible characters in every response?
- Do not assume that every Claude response contains a hidden Unicode watermark. Invisible characters can also be introduced by browsers, rich-text editors and copy-paste. The cleaner reports exactly what it finds, so an already clean result is a real result rather than a failed scan.
- Will this guarantee that AI detectors cannot flag my text?
- No. AI detectors estimate patterns in wording and can produce false positives or false negatives. Removing hidden characters does not change those patterns, and a rewrite cannot guarantee a particular detector score. Treat the result as an editable draft, not proof of authorship.
- What does the rewrite button do?
- It changes sentence structure and wording while preserving the passage's claims, numbers and names. That is the relevant step when a signal lives in token choices rather than hidden characters. Read the result before using it because, unlike the instant cleanup, rewriting changes the prose.
- Does Claude watermark its text?
- Yes. Anthropic documents an imperceptible watermark embedded in text from models released on or after 2 August 2026, across the Claude apps, the API, Claude Code and its cloud partners. Separately, it attaches C2PA signed provenance metadata to .png, .jpg and .svg files.
- What is C2PA?
- The Coalition for Content Provenance and Authenticity standard — digitally signed metadata attached to a file that records how it was made. It travels with the file rather than the words, so copying text out of an image leaves it behind entirely, and no text tool can read or alter it.
- Does it remove em dashes?
- No. An em dash is ordinary punctuation, not a watermark — plenty of people use them, and stripping them would just damage your writing. This only touches characters that are invisible or that impersonate a plain space.
- Is my text uploaded anywhere?
- The initial hidden-character cleanup runs entirely in your browser and uploads nothing. If you press Rewrite, the cleaned text is sent to the rewrite service so it can generate a new version. The two stages are separate so you can use local cleanup without making a network request.
- What is a zero-width space?
- U+200B — a character with no width that renders as nothing at all. It is the most common carrier for hidden data in text, because you can put a long payload in a paragraph and the paragraph looks completely normal.
Remove the mark, then make it yours
Hidden-character cleanup is instant. When the signal lives in the wording, use a structural rewrite, review the result and keep editing until it sounds like you.