Does Claude watermark its text?
Not as a cryptographic signature. Anthropic has not shipped a statistical watermark that proves Claude wrote a given passage. But Claude output does carry a consistent set of hidden Unicode characters — above all the narrow no-break space (U+202F) sitting on both sides of its em dashes. Nobody types that character by hand, so it works as a fingerprint whether or not it was ever intended as one. That is exactly what this tool takes out.
What the “Claude watermark” actually is
The Claude watermark is not one thing — it is two layers, and they need different fixes. The first layer is invisible characters that survive copy-paste into any editor, CMS, or document. The second is visible typography and phrasing habits. This tool handles the first layer completely and the punctuation half of the second; the humanizer further down the page handles the rest.
Here are the characters the remover scans for, what they look like in practice, and why each one matters:
| Character | Where it shows up | Why it matters |
|---|---|---|
| U+202FNarrow no-break space | Wrapped around em dashes in almost every Claude paragraph | The single most common Claude watermark character |
| U+200BZero-width space | Between words or after sentence-ending punctuation | Breaks search, word counts, and string matching |
| U+00A0No-break space | Between numbers, units, and names | Wrecks line wrapping and CSV imports |
| U+200DZero-width joiner | Occasionally between characters mid-word | Invisible, but survives copy-paste everywhere |
| U+FEFFZero-width no-break space (BOM) | At the very start of a copied block | Corrupts file headers and JSON parsing |
| U+E0000–E007FUnicode tag characters | Anywhere; renders as absolutely nothing | Can encode arbitrary hidden data in plain text |
How to remove the Claude watermark
Four steps, about a minute, nothing to install:
- Paste your Claude text. Copy it straight out of the Claude chat window into the box above. Copying is what carries the hidden characters across — retyping the text by hand would leave them behind anyway.
- Click Remove watermark. The scan runs locally and finishes instantly, however long your text is.
- Read the breakdown. You get a count of every marker found, grouped by type, so you can see what was actually hiding in your text rather than just trusting a green checkmark.
- Copy the clean text. The visible wording is byte-for-byte what you pasted. Only the hidden characters are gone.
Where Claude’s hidden characters cause real problems
Most people go looking for this tool because something broke and the cause was invisible. The usual suspects:
- Code that will not run — a zero-width space pasted into a variable name or a string produces a syntax error that looks impossible, because the offending character takes up no space on screen.
- Spreadsheets and CSVs — a no-break space inside a cell silently fails every exact-match lookup, and the cell looks completely normal.
- CMS and email rendering — narrow no-break spaces render as stray boxes or odd gaps in fonts that lack the glyph, so published pages look subtly broken.
- Applicant tracking systems — non-standard whitespace in a résumé can break the keyword matching that decides whether a human ever sees it.
- Search and word counts — Ctrl+F stops finding phrases, and word counts drift, because the text is not made of the characters it appears to be made of.
Removing hidden characters is only half the job
Stripping hidden characters fixes what the text is made of. It does not change how the text reads — and the way Claude writes is a far louder signal than any invisible character. Claude hedges constantly (“It’s worth noting that,” “That said,” “Importantly”), balances both sides of arguments you never asked it to balance, and builds paragraphs in methodical, evenly measured sentences. Strip every hidden character and that rhythm is still there on the page.
So if your goal is text that reads as genuinely human rather than merely clean, run it through the humanizer below after removing the watermark. It rewrites for the patterns that actually give AI writing away: uniform sentence length, stock transitions, hedging, and stiff vocabulary. You can also check the result with our free AI content detector to see which patterns remain.