Fanfiction readers test AI detectors as authors face public callouts
A fan-made AO3 skin flags one Claude artifact, but The Verge reports the method can miss AI use and overstate what it finds.
By Maya Lindqvist · Senior Technology Correspondent
3 min read
A fan-made detector for AI-written fanfiction has intensified disputes inside online fandoms, The Verge reported. The tool, built for Archive of Our Own, matters because readers are using it to flag works and in some cases publicly identify writers accused of using generative AI.
The Verge’s Jess Weatherbed reported that suspicion of tools such as Claude and ChatGPT has been common in creative communities, including fanfiction circles. Readers and writers have traded informal signs they believe point to AI-generated writing, from punctuation habits to ornate prose, but those judgments are uncertain because AI systems are trained on human writing.
The latest effort began June 29, according to The Verge, when an anonymous X account called @heatedrivalryai shared an AO3 skin designed to expose a Claude-related code marker. The account said text pasted directly from Anthropic’s Claude into AO3 can carry a code wrapper labeled “font-claude-response-body,” and that the skin turns a work’s background red when it finds that marker.
The Verge said it tested the skin on AO3 posts created for checking the detector and saw the red-screen warning appear. Weatherbed also reported publishing a Claude-generated short story as a test: the warning appeared when the text was copied straight from Claude into AO3’s editor, then disappeared when the same text reached AO3 through a route that did not preserve the Claude code.
Anthropic did not respond to The Verge’s request to confirm whether the detector works as described, the outlet reported. The Verge said the method appears technically plausible for identifying that specific copied code, but it also described the approach as narrow and prone to misuse.
The limitation is straightforward, according to The Verge: the code marker survives only in certain copy-and-paste conditions. If an author drafts or edits in Google Docs, Microsoft Word or another tool before posting to AO3, the detector may not catch Claude involvement, and writers can remove the code after a work is flagged.
The Verge also reported that the marker does not show how Claude was used. A red screen could indicate a full AI-generated story, or it could reflect a smaller use, such as spell-checking, translation or editing, before text was pasted into AO3.
That distinction has not quieted parts of fandom, The Verge reported. Some readers view any use of generative AI as unacceptable because of concerns about training data scraped from the web and the environmental impact of AI systems.
The Verge said the AO3 skin applies only to one platform and one AI model. The outlet also noted that another person on X claimed to have code that can detect Claude, DeepSeek and some ChatGPT usage, but had not released the tool or explained its method. Google and OpenAI did not respond to The Verge’s questions about whether their systems leave similar detectable traces in generated text.
More reliable disclosure already exists on AO3 through tags, The Verge reported. The site has a “Created Using Generative AI” tag, and some authors use it, but backlash gives writers little reason to volunteer that information.
The Verge reported that at least one writer became caught up in the dispute after someone trusted to edit their fanfiction used Claude. The episode shows the risk of turning a limited technical signal into a broad accusation in a hobby community built around unpaid creative work.
This story draws on original reporting from The Verge.