A fight over AI-written fanfiction has moved from vibes to code, or at least to something that looks like code. The Verge’s Jess Weatherbed reports that fanfic readers and writers have begun using an Archive of Our Own skin posted by an anonymous X account, @heatedrivalryai, to flag works that may have been pasted from Anthropic’s Claude chatbot.
The tool matters because fanfiction communities have long treated authorship as part of the bargain: readers may be there for characters owned by Disney, HBO, or whoever, but they expect the fic itself to come from a human writer. Generative AI has made that trust harder to verify, and the current attempt to police it is already creating its own mess.
How the AO3 skin is supposed to work
According to @heatedrivalryai, Claude can leave behind a specific bit of formatting when a user copies a generated response directly into AO3. The account identified the marker as font-claude-response-body, describing it as a Claude-injected wrapper around the text. The AO3 skin watches for that marker on a page and, when it finds it, changes the page background to red.
That is a blunt instrument. It is checking for a formatting artifact, not proving who wrote a story, how much AI was used, or whether an author edited the text afterward. @heatedrivalryai claimed the marker’s presence shows Claude use “definitively,” according to The Verge. That is the account’s claim, not an independent standard for authorship.
Weatherbed reported that AO3 users have published test works so others can see whether the skin fires. She said the page turned red when she tried the skin on those examples, and she also posted a short Claude-generated story as her own experiment. Those tests show the skin can detect the targeted marker in at least some controlled cases. They do not establish that every unmarked fic is human-written, or that every marked fic should be treated as a bad-faith AI dump.
From punctuation panic to code panic
The Claude skin follows a looser phase of community detective work. The Verge reports that fanfic readers and writers have circulated informal tells for AI-generated writing, including em dashes and ornate prose. Those signs are weak evidence on their own, unless the plan is to accuse half the English-language internet of being a chatbot.
The risk is obvious: a community trying to protect human fanwork may end up turning ordinary stylistic habits, copy-paste residue, or platform quirks into accusations. Weatherbed describes the detection methods now spreading through fanfiction circles as questionable, with any fanfic writer potentially caught in the backlash.
The conflict is also bigger than AO3. The Verge notes that suspicion toward tools such as Claude and ChatGPT has become common in creative communities, especially where writers and artists worry that generative systems are trained on human work and then used to replace it. Fanfiction is now running its own version of that fight, with a CSS-style tripwire standing in for evidence. That is better than guessing from punctuation, barely.
This story draws on original reporting from The Verge.
