Claude watermark checker and remover
Paste your Claude text. Strip hidden HTML, invisible Unicode and copy-paste fingerprints free, then humanize the word-choice pattern in one click.
- Free and unlimited, no account
- Nothing uploaded: checking runs in your browser
What gives AI text away
Markup you cannot see. Copying out of a chat interface carries HTML with it. Claude’s paragraphs arrive wrapped in class names containing “claude”, which land in whatever you paste into. This is what the AO3 detector catches, and it is the reason people get flagged, not their writing style.
Characters that render as nothing. Zero-width spaces, joiners, byte-order marks, non-breaking and exotic spaces. Invisible on screen, obvious to anyone who looks at the bytes, and they survive most copy-paste routes.
The prose itself. Curly quotes, em dashes several to a paragraph, and the even rhythm of sentences all cut to a similar length. Individually weak signals; together, the thing people mean when they say text “reads like AI”.
And the watermark itself. Anthropic’s mark is carried by word choice, which is why stripping characters does nothing to it and why resampling does. Our rewrite pass runs every sentence back through a different model, the one technique Anthropic’s own documentation concedes a mark may not survive.
Paste your text above and see exactly which of these are in it, then clean them in one click. Which Claude models carry the mark →
How to check text for a Claude watermark
Paste the passage into the box above. The check runs as you type, in your browser, with no account and no upload. Every finding comes back with a count and a position in the text, because these are facts about the bytes rather than a probability score you have to trust.
Read the deterministic findings first. HTML class names containing “claude”, zero-width characters, byte-order marks and exotic spaces are either present or they are not. If the class names are there, the text came out of a chat interface without passing through a plain-text editor first. That is the finding worth acting on, and one click removes it.
Treat the stylistic findings as context. Em dashes and curly quotes get counted too, but they prove nothing on their own. We measured ten pre-computer novels and found the rate swings from zero per thousand words in Austen and Stoker to twenty-six in Melville. A signal with that spread cannot accuse anyone of anything.
One thing no checker can do. Anthropic’s statistical watermark is a keyed bias on word choice, and verifying it needs a key that has not been released to anybody. Any tool claiming to detect or confirm it is guessing. We say so plainly, and the detection engine is open source so you can check that claim rather than take our word for it.
Read next
- How to check text for a Claude watermark
- AI watermark detectors: what actually exists
- Does the GitHub Claude watermark remover actually work?
- Does Claude watermark text? (September 2026)
- Best Claude watermark checkers & removers (2026)
- The AO3 AI skin flagged my fic and I did not use AI
- How to paste from Claude without leaving artifacts
- Which Claude models are watermarked? (live list)
Is your model watermarked?
It depends on when the model launched, not which one it is.
Pasted it somewhere already?
What survives a paste depends on where it landed. The fix differs per editor.
Questions
- What exactly does this remove?
- Every trace that is actually detectable in your text. The HTML class names Claude's web interface pastes into your document — the thing the AO3 detector looks for. Zero-width and invisible Unicode characters. Non-breaking and exotic spaces. Smart quotes and em dashes. Then the rewrite pass varies the sentence rhythm and length that make prose read as machine-written.
- Is it free?
- Checking is free, unlimited and needs no account. It runs in your browser, so your text is never uploaded. Cleaning and rewriting cost credits — buy them once you need the rewrite engine. Credits never expire.
- Will my text pass an AI detector?
- Detectors score how predictable your word choices are. The rewrite pass varies sentence length, rhythm and phrasing, which moves exactly the signal those tools measure. The copy-paste artifacts are the part that is hard evidence rather than a probability score, and those come out completely.
- How does the AO3 Claude detector work?
- It is a single CSS rule matching paragraphs whose class name contains 'claude' inside a work's body. That class is only there if the text was pasted straight from Claude without passing through another editor — which is why it catches people. Cleaning the HTML removes it completely.
- Is my text private?
- Checking runs entirely in your browser. The text is never uploaded, so there is nothing for us to store. The rewrite pass sends the passage to our model provider to process it, and we do not keep it afterwards or use it for training.
- How fast is it?
- The check is instant, as you type. A rewrite takes a few seconds for a typical page and scales with length. One credit covers 500 words, and a 20,000-word document is a single job.
- Does Claude watermark its text?
- Yes, on supported models. Anthropic explicitly lists Claude Fable 5.1 and Mythos 5.1 as watermarked at launch. Earlier models remain in a transition period, so we keep a live model-by-model table.
- Does Claude Fable 5.1 watermark text?
- Yes. Fable 5.1 carries Anthropic's statistical word-choice watermark everywhere it is used. Detection is currently in private preview for eligible organisations, so no public checker can verify an individual passage.
- What about that watermark, are you built for it?
- Yes, and it is why the rewrite engine exists. A statistical watermark lives in the word choice itself, so the only thing that touches it is resampling the text through a different model — which is exactly what our rewrite pass does to every sentence. Anthropic's own documentation says marks may not survive heavy editing, paraphrasing or translation. That is the mechanism we are built on, running on your text today.
Claude Watermark is independent and not affiliated with Anthropic.
