Hidden & Invisible Unicode Characters in Text

Hidden Unicode characters are code points that often have no visible ink — zero-width spaces, joiners, bidirectional controls, tag characters, and other non-printing marks. They can ride along when you copy text from editors, websites, or messaging apps. This guide explains what they are and why they matter. It is not a live checker: to inspect a specific paste, use the AI Text Watermark Scanner.

AI Watermark Center’s Claude Unicode study tests whether owned Claude outputs contain observable hidden characters. That protocol is collecting genuine samples and has not published a dataset yet — missing cells are not fabricated. Official Anthropic documentation already states the statistical text watermark is not based on hidden characters; Unicode findings answer a different question.

Why hidden characters are worth understanding

If you paste text into a form, CMS, or compiler, invisible characters can change sorting, search, security filters, or visual layout. Knowing why they appear helps you interpret a scanner report: listing them is useful even when they have nothing to do with watermarks.

What a scanner should not claim

• Presence of a zero-width character is not proof that a named model produced the text.
• Absence of hidden characters is not proof that the text is human-written.
• Stripping hidden characters is not the same as removing a statistical watermark.

How the scanner works

The scanner is deterministic: it walks Unicode code points, classifies them, and shows them. Conservative Cleanup currently removes only a leading BOM candidate. Cleaning re-scans the result so the page can show before and after rather than asserting an unverifiable “undetectable” state.