Your clipboard is carrying characters you cannot see.
Text copied out of AI chat tools often contains invisible Unicode: zero-width spaces, narrow no-break spaces, bidirectional controls. They survive the paste, they are invisible to you, and they are trivially detectable by software. Clipboard Sanitizer removes them the moment you copy.
Large language models emit characters that no one types. The narrow no-break space (U+202F) and the zero-width space (U+200B) turn up routinely in output from AI chat tools, and they travel wherever the text goes.
- AI-detection tools flag your writing on evidence you never knew was in it.
- Invisible characters break code. A zero-width space inside an identifier compiles as a different symbol.
- Diffs and version control show changes on lines that look identical.
- Search, comparison and deduplication silently fail on text that reads the same.
Questions
Does ChatGPT add invisible characters to its output?
Invisible characters are frequently reported in ChatGPT output, most often the narrow no-break space (U+202F) and the zero-width space (U+200B). OpenAI built a text watermarking system and decided not to ship it, so ChatGPT carries no text watermark; what it does carry are characters, and those are removable.
Do Claude, Gemini and other AI tools do the same?
Invisible and unusual Unicode has been reported in output from many AI chat tools. We analysed one sample from each of six assistants and found it in one, Microsoft Copilot, which emitted the non-breaking hyphen (U+2011). No provider has confirmed deliberate invisible-character watermarking. Claude and Gemini do carry a statistical watermark, which is not a character and cannot be cleaned away. The provider pages grade every one of these claims separately.
How do AI detectors detect AI-generated text?
Detectors combine statistical analysis of word choice with simpler signals, and anomalous invisible characters are among the simplest. A zero-width space in the middle of a word is not something a human types, so it is strong evidence about where the text came from. Removing those characters removes that particular signal.
What is a zero-width space?
U+200B is a Unicode character with no width and no visible form. It occupies a position in the text and is preserved by copy and paste, but nothing appears on screen. It is one of several characters designed for typesetting that end up used as invisible markers.
How do I remove invisible characters from text?
You can paste into a plain-text editor, but that strips your formatting and does not catch every character. Clipboard Sanitizer removes them from the clipboard itself, automatically, the moment you copy, so the text is clean wherever you paste it, with formatting intact.
Will it break emoji or non-English text?
No. Zero-width joiners are load-bearing in emoji sequences and in scripts including Devanagari, Bengali, Arabic and Persian, so the app removes them only where they carry no meaning. A family emoji and a Persian word both survive intact; this is the case a naive find-and-replace gets wrong.
Does my text get uploaded anywhere?
No. The app is sandboxed with no network permission and contains no networking code. There is no account, no analytics and no server. Cleaning happens entirely on your Mac and works with the Wi-Fi turned off.
Does it work with passwords?
Copied passwords are never modified. Password managers mark clipboard content as concealed, and the app skips anything carrying that marker, whatever your settings.