CLAUDE WATERMARK REMOVER.
Detect hidden Unicode in Claude text by pasting it into the remover. The report lists every supported mark by name, code point, count and position—before or after you clean.
01 / Detect
Open the detector and remover. Paste Claude text or drop a plain-text file. Nothing is scored until you run a mode.
Careful removes supported hidden characters and returns a finding list. If nothing was present, the report says the result matches the source (apart from NFC notes).
Turn on the Unicode reveal to see invisible characters called out in brackets. That preview is the practical “detector” view for copy-and-paste hygiene.
02 / Coverage
Supported marks include soft hyphen, zero-width space, word joiner, byte order mark, zero-width joiners and non-joiners, and Unicode variation selectors—outside emoji sequences that must keep their joiners and presentation selectors.
Full table: hidden characters reference.
03 / Limits
This is not a Claude authorship classifier. It does not output a probability that text came from Anthropic’s models. Thorough mode rewrites wording; it still does not grade “watermark strength.”
Use detection for hygiene and transparency. Use human review for originality and voice.
04 / Tips
Careful isolates Unicode findings without rewriting sentences. Ideal when you only want to know which invisible characters were present.
On the home workbench, load the built-in sample with hidden marks to see how findings and the reveal toggle behave.
Joiners can be meaningful. The report flags sensitive removals so you can spot emoji or joined-script side effects.
If you paste back into Word, Docs or a CMS, hidden characters can return. Run the detector again on the final plain text.
After you understand the Unicode picture, Thorough can attempt to disrupt the wording pattern—then review meaning carefully.
Mode differences and review questions live on the Claude watermark FAQ.