How to remove Claude's watermark from AO3 (Archive of Our Own)
To post a Claude-assisted fic to AO3 without artifacts, rewrite the prose through a non-watermarking model and strip any stray HTML and invisible characters before pasting into the AO3 editor. AO3 preserves your text, so both the statistical watermark and any leftover chat-interface characters would otherwise carry into the posted work.
What AO3 (Archive of Our Own) does with Claude's watermark
AO3's rich-text and HTML editors keep what you paste, including zero-width characters, smart quotes, and stray span or class markup copied from a web chat. AO3 itself adds no watermark. Cleaning matters here because posted works are public and permanent, and the community is sensitive to AI provenance.
Steps to clean it
- 1Paste your Claude text into ClaudeRewrite. It rewrites the wording through a non-watermarking model (DeepSeek V4 Flash), breaking the statistical pattern while preserving every fact, number, name and link, then strips any leftover invisible characters.
- 2Paste the cleaned prose into AO3's editor. If you used the HTML editor, check for and remove any leftover span/class tags copied from a chat interface.
- 3Preview the work to confirm formatting, then post.
FAQ
Why does Claude text show weird characters on AO3?
Text copied from a chat interface can carry invisible Unicode and smart quotes. These are not the watermark, but AO3 preserves them. ClaudeRewrite strips them along with breaking the watermark.
Can AO3 readers detect AI writing?
There is no public watermark detector today. The visible tells are stray characters and AI phrasing, both of which cleaning addresses.
Remember: Claude's watermark is statistical — it lives in the word choices, so it travels with your text into AO3 (Archive of Our Own) no matter how you paste. There is no public detector today, but published text stays checkable once Anthropic's detection API ships. See which models are marked →