Claude Watermark
Find out what actually gives your AI text away
Anthropic is watermarking Claude's text. Tools selling "watermark removal" strip invisible Unicode — which cannot touch that mark, because it is distributional: a secret key biases word choice, so the watermark IS the wording. Paste text and see, in your browser, every fingerprint a detector actually finds: the HTML classes copy-paste leaves behind (what the AO3 Claude detector checks), zero-width characters, odd spaces, stylistic tells. Nothing is uploaded. One button then cleans those artifacts and rewrites the prose in two passes. Free to check. 10 credits on signup, no card.

Anthropic is watermarking Claude's text. Tools selling "watermark removal" strip invisible Unicode — which cannot touch that mark, because it is distributional: a secret key biases word choice, so the watermark IS the wording. Paste text and see, in your browser, every fingerprint a detector actually finds: the HTML classes copy-paste leaves behind (what the AO3 Claude detector checks), zero-width characters, odd spaces, stylistic tells. Nothing is uploaded. One button then cleans those artifacts and rewrites the prose in two passes. Free to check. 10 credits on signup, no card.
1 comment
Every tool ranking for "Claude watermark remover" strips zero-width Unicode, which can't touch a distributional mark. Against my own interest: nothing is watermarked yet, and no detector exists.