If you have ever copied a Claude reply into Google Docs, Word, or a Git commit and suddenly seen odd spacing, broken search, or a diff full of “empty” changes, you are not imagining it. Some large language models leave behind invisible Unicode characters and odd typographic marks. They do not show up as a visible watermark stamp. They sit in the text as zero-width spaces, soft hyphens, unusual dashes, or lookalike punctuation.
What usually goes wrong
Invisible characters break everyday workflows in quiet ways:
- Find-and-replace misses words because a zero-width character sits inside them.
- - Grammar tools score the same paragraph differently after a paste.
- - Code review diffs light up lines that look identical on screen.
- - PDF or CMS exports insert strange line breaks.
None of this means Claude “branded” your file with a cryptographic watermark you can see. It means the draft may still carry leftover formatting artifacts from generation or from the copy path through the browser.
A practical cleanup checklist
- Paste into a plain-text field first, not straight into a design tool.
- 2. Scan for em dashes, curly quotes, and soft hyphens if your style guide bans them.
- 3. Re-check search inside the document after cleanup.
- 4. For Git, stage the file and review the diff before committing.
If your goal is readable, searchable, commit-friendly prose, removing invisible leftovers is usually enough. If your goal is academic integrity or model-detection evasion, that is a different problem: detectors score style and probability, not just Unicode. Clean the artifacts for craft and tooling. Keep the writing honest for the audience.
For a fast, local cleanup pass on publish-ready text, try Claude Watermark Remover
Sign in to leave a comment.