Claude just started watermarking everything it writes, and people are cancelling their subscriptions over it

Claude has started watermarking the text it writes, and people are cancelling their subscriptions over it.

I think they're solving the wrong problem.

Europe's new AI law tells every AI company to mark AI-written text, but it doesn't say how. Google already stamps a pattern into Gemini's output, and Anthropic has now built its own version into Claude itself. The watermark lives in the word choices the model makes, so pasting it through Notepad does nothing. There are no hidden characters to strip out.

Within days, free tools appeared promising to scrub the pattern out by rewording your text through a second AI until it breaks.

I'd leave them alone. Anthropic hasn't released the detector yet, so nobody selling a scrubber can prove their tool works. And swapping every sentence for near-synonyms flattens your writing into generic mush. You're trading your voice for camouflage that might not hold.

Nobody is running your newsletter or your proposals through a detector. Your customers care whether it sounds like you and whether it's true, and no watermark changes either answer.

Take the AI draft, spend ten minutes editing it into your own voice, and put your name on it. That was the job before watermarks, and it still is.