Remove what can be removed
Three stages, each labelled with its own confidence level. Stage one detects invisible characters and changes nothing. Stage two removes them only when you ask, then rescans the output so you can watch the count reach zero — that step is fully verifiable.
Stage three optionally proposes a prose rewrite and reports the percentage of wording it would change, because a substantial rewrite is the only thing that genuinely disrupts a statistical watermark. Nothing replaces your text until you accept it.
What this tool cannot do: confirm that a statistical watermark is gone. No public detector for Claude's text watermark exists, and Google's own SynthID Detector portal accepts image, video and audio but not text. Any tool reporting your text is now clean is describing a capability that does not exist publicly.
Frequently asked questions
Can this remove Claude's watermark?
It removes every invisible character and can rewrite the prose, which is what genuinely weakens a statistical watermark. It cannot confirm removal, because the detector is private to Anthropic. Any tool claiming confirmation is claiming a capability that does not exist publicly.
Is my text uploaded anywhere?
No. All processing runs as JavaScript in your browser. There is no server endpoint to receive text, so nothing is uploaded, stored or logged. You can verify this in your browser's network tab.
Why did it find zero invisible characters in my Claude output?
That is the correct result. Anthropic states that nothing is added to the text and there are no hidden characters. Claude's watermark is a pattern in word choice, so a character scan will correctly find nothing.