How to remove hidden characters from ChatGPT and other AI text

Text copied from ChatGPT, Claude, Gemini or Copilot can carry characters you cannot see. Paste it into Gren and they are removed as the text lands, with a short note of how many there were. It is free, needs no account, and your text never leaves your browser.

Updated 10 October 2026 · 5 minute read

What the hidden characters are

None of these show on screen, but each one is a real character in the text.

• Zero-width spaces and joiners (U+200B, U+200C, U+200D, U+2060) and the byte order mark (U+FEFF). They slip in when you copy from a chat window.
• Odd-width spaces, such as the narrow no-break space (U+202F) and the no-break space (U+00A0). In April 2025, OpenAI's o3 and o4-mini models put narrow no-break spaces where normal spaces belonged. OpenAI said it was a quirk of training, not a watermark.
• Soft and non-breaking hyphens (U+00AD, U+2011). One is invisible and the other looks like a normal hyphen, but both break search and spell-check.
• Direction marks (U+200E, U+200F and their relatives), which tell text which way to run.
• ChatGPT citation markers: invisible characters wrapped around cited phrases in answers that searched the web. Some apps also leave visible scraps such as citeturn0search0.
• Hidden tag characters (U+E0000 to U+E007F), which can carry whole sentences nobody can see.
• Look-alike and styled letters: a Cyrillic "о" in place of a Latin "o", or the bold and italic letters people paste into LinkedIn posts, such as "𝗕𝗼𝗹𝗱".

Why they cause trouble

• Search and find and replace miss words with a hidden character inside them.
• Spell-check flags words that look perfectly spelt.
• Code, formulas and spreadsheet cells stop working, often with no clue why.
• Word and character counts come out wrong, which matters when a post has a limit.
• Hidden text can carry instructions meant for another AI tool rather than for your reader.

How to remove them with Gren

1. Copy the text from the AI tool as you normally would.
2. Open gren.app and paste into a note.
3. Hidden characters are removed as the text lands. So are Markdown symbols such as ** and ##, dashes become commas and curly quotes become straight ones. A message says what was removed.
4. Changed your mind? Press Undo clean in that message and you get exactly what you copied.

For text that is already in a note, press Cmd K (Ctrl K on Windows) and choose Clean AI text. It works on a selection or the whole note, and Cmd Z takes it back. To paste text exactly as copied, switch off Clean AI text when pasting in the same list.

Gren is careful with what should stay. Your own spacing and indents, emoji (skin tones and flags included), other languages, code, web addresses and file paths are left as they are. A hidden character between two words never leaves them glued together or with a double space.

Why "paste as plain text" is not enough

Pasting as plain text removes fonts, colours and links. It does not remove these characters, because they are text, so they come through any plain text paste unchanged.

What cleaning cannot do

Removing hidden characters does not make AI writing undetectable. AI detectors judge the wording, not hidden characters, and Turnitin, for one, swaps look-alike letters back and flags hidden text before it checks. The watermarks that do exist, such as Google's SynthID, are built into word choice, so no character cleaner can remove them. OpenAI has said ChatGPT does not add a hidden watermark to its text. If you are writing for a course or an employer, their rules on AI use are what count.