AI text watermark remover for hidden Unicode

What this page checks

Hidden characters in pasted AI text

Text copied from ChatGPT, Claude, Claude Code, Gemini, Microsoft Copilot, Grok, or another AI tool can be checked here. InvisiChar treats the pasted text the same regardless of where it came from: it identifies only characters in its documented Unicode registry, explains them in context, and lets you review supported changes.

The checker can find supported zero-width and format characters. It can also flag U+00A0 NO-BREAK SPACE, U+2007 FIGURE SPACE, U+2009 THIN SPACE, and U+202F NARROW NO-BREAK SPACE for optional replacement with one ordinary U+0020 SPACE. It never changes text merely because it came from an AI product.

How to check and clean the text

1. Paste the text into Inspect and choose Inspect text.
2. Read each finding’s code point, context, risk, and legitimate use.
3. Open Clean and select only the removals or fixed space replacements you want.
4. Apply the selected changes, review the diff, then copy the result.

Everything in the text area, including findings and output, stays in this browser. Read the safe-removal guide before changing unfamiliar characters.

What a finding means

A supported hidden character is evidence that the code point exists in this pasted text. It does not identify which application inserted it. Ordinary software, websites, editors, writing systems, and intentional formatting can use the same Unicode characters.

“AI watermark” can also refer to provenance metadata, a signal embedded in media, or a statistical pattern in generated language. Those are different mechanisms. This page does not detect whether text was written by AI, verify an official watermark, remove a statistical text signal, rewrite AI-sounding prose, strip file metadata, process image or audio watermarks, or help bypass an AI detector.