What this tool cannot tell you

It is not Anthropic’s official detector

Anthropic has not published the technical detection mechanism referenced in its current guidance. This tool does not imitate that mechanism, call an undisclosed endpoint, or assign a Claude watermark confidence score.

Unicode findings do not prove authorship

Hidden or unusual characters can come from word processors, websites, copy-and-paste, accessibility tools, multilingual typography, source code, and many other systems. A finding only proves that the specific code point occurs at the reported location.

A clean report does not prove no watermark

The inspector covers a defined set of invisible Unicode, spacing, control, direction, variation, and noncharacter code points. Claude’s documented text watermark is model-level. Removing Unicode artifacts does not establish that a statistical signal is absent.

Cleaning can change rendering

Nothing is selected by default. Items labeled “Review” may affect word wrapping, emoji presentation, ligatures, bidirectional text, mathematical notation, or complex script shaping. Inspect the diff and test the result in its destination.

Rewriting has no guarantee

Substantive editing can weaken detectability according to Anthropic’s stated limitations, but no independent tool can convert that observation into a guarantee for a particular passage. Rewrite mode is unconfigured and produces no simulated output.

No compliance bypass

The tool is for transparent text hygiene, publishing, debugging, and informed editing. It should not be used to misrepresent authorship, evade required disclosure, or bypass institutional or platform rules.