Claude Watermark

You can't use "Claude" in the name, ban incoming.

This is such a great response to their updates :D

You ran ten pre-computer novels through the checker to test the em dash theory. Moby Dick hits twenty-six per thousand words, Austen none. Congrats on shipping something that leads with what it can't do.

I don't see any issue with cloud watermarking, but anyway, good niche findings! Good luck!

Cool approach, and agreed that you can't detect Anthropic's watermark without their keys. But on removal I wouldn't fully agree: you don't need to detect a watermark to remove it. There's research showing these watermarks mostly don't survive paraphrasing or translating back and forth.

Hey Product Hunt 👋 I built this after the Claude watermark news in August, when the internet filled up with tools promising to remove a watermark nobody outside Anthropic can actually detect. This does the part that is real. Copying out of a chat interface carries things with it: HTML class names, zero-width characters, exotic spaces, typography. Paste anything and you get every one of them with a count and a position, because those are facts about the bytes rather than a probability score. One click strips them. Two things I would rather say up front than bury. First, it does NOT detect Anthropic's statistical watermark. Nobody outside Anthropic can, because verification needs a key that has not been released. Any tool claiming otherwise is guessing. Second, the em dash is not the tell people think it is. I ran ten pre-computer novels through the checker: Melville uses 26 per thousand words in Moby Dick, while Austen and Stoker use none at all. A signal that swings from 0 to 26 across human authors cannot accuse anyone of anything. The detection engine is MIT open source, so you can check both of those claims instead of trusting me. Free and unlimited to check, runs in your browser, nothing uploaded. I would genuinely like to hear where it gets things wrong.