In brief
A reliable independent Claude watermark checker cannot be implemented until Anthropic publishes its technical detection guidance.
The local checker can still report supported hidden Unicode formatting characters by exact code point. It does not analyze style, guess authorship, or return a fabricated confidence score.
Detection availability check
Check Claude watermark status
Paste text to scan for the supported Unicode formatting characters documented by this site. Your text stays in this browser and is not uploaded.
Supported character scan complete
Code-point findings
Official Claude watermark verification unavailable. This scan covers disclosed Unicode formatting rules only; it cannot prove Claude authorship or the presence or absence of Anthropic's unpublished watermark.
Process This Text InsteadCan Claude watermarks currently be checked?
Anthropic says it is working to enable users and third parties to detect supported Claude text watermarks and file provenance metadata. The public article says technical detection details are forthcoming.
Without that specification or an official detector, this site cannot reliably determine whether pasted text carries a Claude mark. Visual style and generic AI-writing signals are not substitutes for watermark detection.
What will this checker do when documentation becomes available?
Any future checker must be based on public, testable guidance from Anthropic. Before release, it would need defined inputs, documented result meanings, known error conditions, privacy review, and tests against authoritative examples.
The interface is deliberately separate from the detection engine so a verified method can replace the current status-only behavior without changing the site’s privacy rule: pasted text must remain local unless users are clearly told otherwise.
What does positive detection actually mean?
Anthropic says a supported positive mark would indicate that content may have been processed by Claude. It would not, by itself, establish complete provenance or prove that Claude originated every idea or sentence.
Claude can proofread, translate, summarize, or transform material supplied by a person. Content can also be excerpted, edited, or combined after Claude processes it.
What the supported character scan reports
Standard inspection covers exotic spaces U+00A0, U+1680, U+2000–U+200A, U+202F, U+205F, and U+3000; U+2028–U+2029 separators; U+00AD, U+200B, U+2060, U+FEFF; and U+E0001–U+E007F tag characters.
Broad inspection adds U+034F, U+061C, U+200C–U+200F, U+202A–U+202E, U+2061–U+2064, U+2066–U+206F, and U+FFF9–U+FFFB. The result lists only code points found in the supplied text and never labels text “watermark clean.”
Why negative detection would not prove human authorship
Anthropic lists several reasons Claude-processed content might have no detectable mark: it may come from a model without marking support, be heavily edited or translated, be mixed with other writing, or be too short for a reliable signal.
A negative check would therefore mean only that a supported mark was not detected under that detector’s conditions. It would not be a certificate that no AI system participated.
Claude Watermark Checker FAQ
Does this checker analyze my pasted text?
It inspects only the documented Unicode formatting categories selected above. It does not inspect style, calculate an authorship score, run Anthropic’s detector, or upload the text.
What does a zero finding mean?
It means no characters in the selected supported coverage were found. It does not prove that Claude did not process the text or that Anthropic’s unpublished watermark is absent.
Will my text be carried to the remover?
No. The Process This Text Instead link opens the homepage tool without placing your text in a URL or browser storage. Paste it again if you choose to process it.
When will reliable detection be available?
Anthropic says technical documentation is forthcoming but has not provided a date in the currently cited article.