A straight answer first: zero-width characters are not an official OpenAI watermark. OpenAI has not announced a text watermark in ChatGPT. In April 2025 people found narrow no-break spaces in some ChatGPT answers, and OpenAI described them as a side effect of training, not a mark. Invisible characters also arrive through web pages, PDFs, Word and chat apps.
So this tool does not promise to defeat a detector. It does one checkable thing: it shows what is hidden in your text and removes it. That fixes broken searches, odd line wraps, strange spacing and diff noise. AI detectors score writing style, so cleaning characters will not change how they rate a text.
What it removes: zero-width spaces, joiners and non-joiners, word joiners, byte order marks, soft hyphens, narrow and ordinary no-break spaces, other odd spaces, bidirectional marks, variation selectors and tag characters. Emoji sequences are left intact unless you turn that off.
How to use it
- Paste the text into the box.
- Look at the Found panel. Each hidden character is a small labelled tag, and the counts are listed by type.
- Press Copy clean text.
Questions
Does ChatGPT put a watermark in text?
OpenAI has not announced one. Hidden characters in pasted text mostly come from formatting and copy paste. Plaincopy shows you what is in your text; it cannot tell you where it came from.
Will removing them hide that text came from AI?
No. Detectors look at wording patterns, not invisible characters. We do not claim otherwise.
Is my text uploaded?
No. Cleaning happens in your browser tab.