What the text cleaner actually checks

Paste text or open a UTF-8 .txt file in Free AI Wiper. The browser identifies eleven supported categories: zero-width spaces, word joiners, soft hyphens, bidirectional controls, embedded byte-order marks, Unicode tags, noncharacters, reserved invisible characters, deprecated format controls, private-use characters and unusual spaces. Findings include the Unicode code point, line and column so you can decide whether a change is appropriate.

Hidden Unicode characters are not necessarily AI metadata. The text cleaner does not judge who wrote a passage. It cannot detect statistical AI watermarks or signals that are not present as one of those characters. Many invisible characters have legitimate uses, especially in multilingual text. The cleaner deliberately leaves common emoji and script joiners alone.

Complete black-flag tag sequences, variation selectors, script joiners, script fillers and invisible mathematical operators stay intact. Space normalization replaces selected unusual spaces with ordinary spaces instead of joining words. Bidirectional controls, private-use characters and unusual spaces are reported but not selected by default: they may carry intentional layout or custom-font meaning. There is no automatic rewriting, compatibility normalization or conversion of lookalike letters.

Inspect before removing

  1. Open the text cleaner and paste text, or choose a UTF-8 .txt file up to 1 MiB.
  2. Review the code points and positions. Pasted text is limited to one million UTF-16 code units and inspection stops above 20,000 findings to keep reports bounded.
  3. Select the trace kinds you want to remove or spaces to normalize, then copy or download the cleaned text. The result counts removed characters and normalized spaces separately.

If a character is intentional, keep it. Removing a bidirectional control, for instance, may change how nearby text displays. The step-by-step guide shows how to investigate a finding before editing it.

For OpenAI textGrain and Claude’s word-choice signals, read how ChatGPT and Claude text watermarks differ from hidden characters. This cleaner checks only the supported characters listed above; it does not detect or remove those statistical watermarks.