The families the inspector reports
Each finding names the code point and its Unicode name, so nothing is guessed.
- Zero-width space, non-joiner, joiner and word joiner (U+200B–U+200D, U+2060).
- Byte order mark used mid-document (U+FEFF).
- Bidi controls: marks, embeddings, overrides and isolates.
- Tag characters (U+E0000 block), which can carry an entire hidden message.
- Variation selectors, private-use code points, noncharacters and reserved default-ignorable code points.
- Space homoglyphs such as no-break space, narrow no-break space and ideographic space.
Why they matter in practice
A zero-width space inside an identifier makes a search fail without any visible reason. A narrow no-break space in a CSV column breaks a numeric parse. A bidi override can make a filename display in a different order than it is stored.
This is text hygiene. It is not a way to change how a detector scores writing, and MetadataRemover.cc does not claim that it is.
What is deliberately preserved
A naive strip destroys legitimate multilingual text, so the cleaner keeps invisible characters that are doing real work.
- Zero-width joiners and non-joiners between letters of the same script, where they are orthographic.
- Emoji glue: the joiner inside a composed emoji and the variation selectors that choose its presentation.
- Complete regional-subdivision flag tag sequences.
- Mongolian free variation selectors, Khmer inherent vowels and Hangul fillers after their own letters.
- Directional marks and isolates, and correctly paired embeddings, which mixed right-to-left text needs.
How this version handles text
Unusual spaces are normalised to ordinary spaces by default. Characters needed for supported writing systems and emoji are preserved in context. This page does not offer advanced switches for bidirectional controls or look-alike letters. Review your cleaned text before using it, especially where spacing has a typographic or technical purpose.
Frequently asked questions
How do I find hidden characters before removing them?
Paste your text into the inspector and run a check. Findings show supported code points and whether cleaning removes or preserves them. Review that list first, then clean and copy the result. MetadataRemover.cc does not rewrite your sentences.
Can I clean hidden characters in copied ChatGPT text?
Yes, if the pasted text contains characters the tool supports. Hidden characters can come from many sources, not just ChatGPT. Removing them is not the same as removing a statistical AI watermark, and it does not prove who wrote the text.