What Claude watermark claims can and cannot do
Claude can leave a statistical pattern in the words it chooses. A thorough paraphrase can weaken that signal. That is not a guarantee, and this page is not a guide to beating detectors.
What the watermark is
When Claude writes, it leaves a statistical pattern in token choice. Anthropic’s method is a version of Google DeepMind’s SynthID-Text. Nothing is inserted into the file. There are no hidden characters. You cannot see it on the page. Anthropic described it on August 14, 2026.
Anthropic: How Claude’s text watermark works; How Claude marks AI-generated content.
What a paraphrase can do
A thorough rewrite can weaken the signal. Anthropic’s own note is that light edits probably will not clear it, while replacing the words can. Dense facts and code have fewer places for a mark to live. If a mark is later found, it only means Claude was likely involved at some point. It does not say who wrote the piece.
Anthropic: How Claude’s text watermark works.
What those claims cannot do
We do not promise the mark is gone. We do not call the result undetectable. Short passages do not give you much to go on. A light synonym pass is not a thorough rewrite. Mixing a few original sentences back in can keep a signal that a full pass would have moved.
This is not a detector-bypass guide. We do not publish settings, tricks, or success rates for clearing a mark.
What Phravv does not do
Phravv does not detect the mark. No badge, no score, and we are not building a detector. If Anthropic opens a public detection API, we might use it behind the scenes and add a sentence to the FAQ. The rewriter will not depend on it.
Phravv never rewrites with Claude. If Claude rewrites the text, it can put the same kind of mark back in.
How to read this
Use the rewriter because you want a second pass that keeps the meaning. Read Anthropic’s own notes if you care about the mark. Treat any “guaranteed removal” claim — ours or anyone else’s — as marketing, not evidence.