The AO3 AI skin flagged my fic and I did not use AI
By Claude Watermark Research · Updated
The AO3 Claude skin does not analyse writing style, so it does not produce style-based false positives the way tools like GPTZero do. It checks for one HTML artifact that can only be present if text was pasted directly from Claude's interface. However, a flag proves the text passed through Claude — not that Claude wrote it. Running your own prose through Claude for proofreading, translation or formatting produces exactly the same artifact as generating it from scratch.
What a flag actually means
It means the HTML of your posted work contains a class name with “claude” in it. That attribute cannot appear by accident. It can only get there by pasting from Claude's web interface into AO3's rich-text editor without an intermediate step.
So the detection is accurate about the mechanism. What it is not accurate about is the conclusion people draw from it.
The gap between the artifact and the accusation
Anthropic itself makes this point about its watermark, and it applies just as well here: text you wrote and asked Claude to proofread carries the same trace as text Claude wrote outright. So does text you translated, reformatted, or asked for a spelling pass on.
A reader seeing the banner has no way to distinguish “Claude wrote this fic” from “the author fixed their commas in Claude”. Both produce an identical flag.
If your work has been flagged
Remove the artifact — our guide on removing Claude's HTML covers both the quick and the precise method. That stops the banner appearing.
If you did use AI and are comfortable saying so, tagging it is the option a lot of the community actively asked for. In the largest thread on this, the top comment argued exactly that: readers who do not mind can filter in, readers who do can filter out, and nobody is deceived.
If you did not use AI in any form and are still flagged, check whether anything in your posting workflow routes through Claude — a browser extension, a grammar tool, a co-author's draft.
Questions
- Can the skin flag work that never touched Claude?
- Only if someone deliberately typed the class name into their HTML, which effectively does not happen. Unlike style-based detectors, this one does not guess.
- Is this the same as an AI detector?
- No. AI detectors estimate from writing style and are wrong often, particularly for non-native English writers and technical prose. This checks for a specific artifact, and is right about that artifact — the disagreement is about what it implies.
Check your own text. Free, unlimited, no account, and it runs in your browser so nothing is uploaded.
Open the checker
