Detect and strip zero-width spaces, U+202F narrow spaces, BOMs and other hidden unicode from text — including the characters reported as ChatGPT watermarks.
Stays on your deviceFree
Result
LLM output and copy-pasted web text often carry invisible unicode: zero-width spaces, word joiners, narrow no-break spaces (U+202F — a known GPT-5 artifact), BOMs and directional marks. They break search, diffs, character counts and code. This page detects them, shows you exactly what was found and where, and strips them.
FAQ
The characters reported as watermarks (like U+202F in o3/o4-mini output) are exactly what this tool removes. Researchers consider them training artifacts rather than deliberate watermarks — either way, they're gone after cleaning.
Removal targets zero-width and directional characters only. Note that right-to-left text (Arabic, Hebrew) can legitimately use directional marks — check the detection panel before cleaning mixed-direction text.
No. Detection and cleaning run 100% locally — nothing is uploaded.
How it works & more
LLM output and copy-pasted web text often carry invisible unicode: zero-width spaces, word joiners, narrow no-break spaces (U+202F — a known GPT-5 artifact), BOMs and directional marks. They break search, diffs, character counts and code. This page detects them, shows you exactly what was found and where, and strips them.
FAQ
The characters reported as watermarks (like U+202F in o3/o4-mini output) are exactly what this tool removes. Researchers consider them training artifacts rather than deliberate watermarks — either way, they're gone after cleaning.
Removal targets zero-width and directional characters only. Note that right-to-left text (Arabic, Hebrew) can legitimately use directional marks — check the detection panel before cleaning mixed-direction text.
No. Detection and cleaning run 100% locally — nothing is uploaded.