FAQ
How the rewriter works, and an honest note on Claude’s statistical text watermark.
What is Phravv?
A paraphrase tool at phravv.com. Paste a draft, pick a strength, and get a second pass that keeps the meaning and changes the wording. The facts stay. The sentences move.
How do Light, Balanced, and Aggressive differ?
They decide how far the sentences move. Light keeps sentence order and leans on synonyms. Balanced rephrases most sentences and keeps the argument order. Aggressive is allowed new sentence boundaries and voice, with the same facts. They are craft controls, not stealth settings.
What does Phravv leave alone?
Numbers, names, dates, quotations, URLs, heading lines, and fenced code. It is a constrained rewriter, not a new draft from a prompt.
Do you keep what I paste?
The text is sent only to produce the rewrite. We don’t keep a server-side history, and we don’t sell what you paste. Recent pairs may stay on this device in your browser until they’re overwritten or you clear site data.
How long a draft can I paste?
About 4,000 characters, enough for a short document. Longer pieces work in sections.
What is Claude’s text watermark?
When Claude writes, it leaves a statistical pattern in the words it chooses. Anthropic’s method is a version of Google DeepMind’s SynthID-Text. Nothing is inserted into the file, and there are no hidden characters. You can’t 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 does a paraphrase do to it?
A thorough rewrite can weaken the signal. Anthropic’s own note is that light edits probably won’t clear it, while replacing the words can. We don’t promise that, and we don’t call the result undetectable. Short passages don’t give you much to go on. Dense facts and code have fewer places for a mark to live in the first place. If a mark is found, it only means Claude was likely involved at some point. It doesn’t say who wrote the piece.
Anthropic: How Claude’s text watermark works.
Why doesn’t Phravv rewrite with Claude?
If Claude rewrites the text, it can put the same kind of mark back in. Phravv never calls Anthropic models.
Do you detect the mark?
No. No badge, no score, and we aren’t building a detector. If Anthropic opens a public detection API, we might use it behind the scenes and add a sentence here. The rewriter won’t depend on it. We don’t sell a way around detectors, and we don’t pitch this at students trying to beat school checkers.