Built by Pluto.Studio · find and test independent Android apps on IndieHangar
AI Watermarks

Is my text watermarked?

This tool finds hidden characters in text — zero-width characters, unusual spaces, and letters from other alphabets that look like Latin ones. It runs entirely in your browser; nothing is uploaded. It cannot tell you whether text carries Claude's watermark, because that mark is a statistical pattern in word choice rather than an added character. Only Anthropic can check for it, through a detection API that has been in private preview since 1 September 2026 for organisations that EU law obliges to verify marking[1]. No public tool can tell you that either[1].

No upload, no account, no logging.

Check an image for Content Credentials

Claude attaches signed C2PA provenance metadata to the .svg, .png and .jpg files it generates[1][10]. This checks whether a file carries such a manifest. The file is read locally and not uploaded.

Drop an image here, or click to choose one

What the scanner looks for

CodepointNameNotes
U+200BZero-width spaceInvisible. A genuine hiding place for markers.
U+200CZero-width non-joinerInvisible. Legitimate in Persian, Arabic and Indic scripts.
U+200DZero-width joinerInvisible. Legitimate inside emoji sequences.
U+FEFFZero-width no-break space / BOMInvisible. Often a file-encoding artefact.
U+00ADSoft hyphenInvisible unless the line wraps.
U+2060Word joinerInvisible.
U+180EMongolian vowel separatorInvisible. Almost never legitimate in English.
U+00A0Non-breaking spaceLooks like a space. Common from word processors.
U+2007Figure spaceLooks like a space.
U+2008Punctuation spaceLooks like a space.
U+2009Thin spaceLooks like a narrow space.
U+202FNarrow no-break spaceLooks like a narrow space.
U+2028Line separatorInvisible line break.
U+2029Paragraph separatorInvisible paragraph break.

Plus Cyrillic and Greek letters that render identically to Latin ones — а, е, о, р, с and their capitals. These are how some systems hide markers in plain sight. They are also completely normal in genuinely multilingual text, so a hit is a question rather than a verdict.

How to read a clean result

A clean result means no hidden characters. It does not mean the text was written by a human, and it does not mean the text is unmarked. Claude's watermark lives in the word choices themselves[1], and Google's SynthID Text works the same way, by adjusting token probability scores[3]. Neither leaves a character you can find, and only Google has published a working detector[4].

If you are here because you were accused of something, the page on false accusations will help more than this tool will.

Sources

  1. How Claude marks AI-generated content primaryAnthropic, 2026-08, updated 2026-09-16
  2. SynthID primaryGoogle DeepMind, 2026
  3. google-deepmind/synthid-text primaryGoogle DeepMind, 2024-10
  4. How Claude's text watermark works primaryAnthropic, 2026-08-14, updated 2026-09-01

Last verified against primary sources: