Checker
How to check Claude text for watermarks
Searching “check Claude watermark” usually means two different jobs. This page covers the one you can actually do today: inspect the pasted string for visible-in-source traces. Anthropic has not shipped a public detector for the invisible statistical mark in word choice.
What this checker looks at
Checking runs in your browser as you paste. It reports:
- HTML class names containing “claude” — common when you copy from Claude artifacts or rich HTML.
- Zero-width or invisible characters — joiners and BOM marks that survive copy/paste into some editors.
- Non-breaking or exotic spaces — including nbsp and other Unicode spaces that look like normal gaps.
- Em dashes and curly quotes — stylistic only. They are not proof of Claude.
“Clean traces” strips the first three. It does not rewrite sentences, and it does not claim to erase Anthropic’s sampling watermark. That split is spelled out in clean traces vs rewrite.
What you cannot check (yet)
Claude’s main text watermark is a statistical bias in token choice. A verifier would score whether a passage matches that secret pattern. Anthropic has discussed detection for users and third parties; a public consumer detector is not the same as the mark already being embedded in outputs.
Until that exists, treat “check” as: look for inspectable leftovers, then decide whether you also need a meaning-preserving rewrite. Do not trust GPTZero, Turnitin, or ChatGPT as a Claude-watermark oracle — see GPTZero, Turnitin, and ChatGPT.
How to run the check
- Paste the Claude output. Copy the passage — including HTML from artifacts if that’s what you have — into the checker on the homepage.
- Read deterministic vs stylistic findings. Deterministic hits are HTML class names containing “claude”, zero-width characters, and exotic/nbsp spaces. Em dashes and curly quotes are stylistic, not proof.
- Clean traces, then decide on rewrite. Clean traces for free in-browser. The statistical mark lives in token choice; neutralizing that takes a paid rewrite with a non-Claude model.
How to read the result
- Deterministic hits: something in the paste is inspectable. Cleaning is free and local.
- Stylistic notes only: punctuation taste, not a watermark. Ignore them if you like em dashes.
- No findings: the string looks clean in-source. The statistical mark can still be present in the wording. Checking cannot prove “not Claude.”
FAQ
Can I check Claude’s statistical watermark in this tool?
No. Anthropic has not shipped a public Claude watermark detector. This checker looks for inspectable traces in the pasted string — HTML class names containing “claude”, zero-width characters, and exotic or non-breaking spaces.
Are em dashes or curly quotes proof of Claude?
No. Those are stylistic. Lots of human writing uses them. The checker flags them as optional notes, not as a watermark hit.
Does checking upload my essay?
Checking runs in your browser. Nothing is sent to the API until you start a paid rewrite.
Related guides
- All Claude watermark guides
- Claude watermark August 2026 news
- Remove Claude watermark
- Clean traces vs rewrite
- How Claude watermarks work
- Claude AI text watermark
- Claude artifacts HTML classes
- GPTZero, Turnitin, and ChatGPT
- Claude watermarks on essays
- Rewrite credits explained
- Anthropic watermark (EU AI Act)
- Claude C2PA provenance
- Remover vs paraphrase tools
Not affiliated with Anthropic. This checker is not a public Claude watermark detector. Updated August 30, 2026.