Claude detector divides AO3 fanfiction community and readers

An anonymous X account called @heatedrivalryai posted a skin on June 29 that flags a Claude-specific markup on Archive of Our Own, turning flagged pages bright red when the injected code is present. The skin looks for a Claude-injected wrapper, which the account described as "font-claude-response-body," and the author who tested the skin found the red background appears only when text is pasted directly from Claude into AO3.

What does the AO3 Claude detector do?

The skin detects a specific HTML wrapper that appears when text is pasted directly from Anthropic’s Claude, and it visually marks pages by changing the background to red. The anonymous account behind the tool said, "When a Claude-generated response is pasted directly into AO3 from Claude, the text is wrapped by a Claude-injected code 'font-claude-response-body.'" The Verge writer tested the skin, posting a Claude-generated short story and observing the red screen when pasting straight from Claude, and the red disappeared when the identical text was pasted after being moved through other editors.

How reliable and comprehensive is this method?

The detector can reliably find that specific artifact only in narrow circumstances, but it misses most other cases. The code wrapping is preserved only if content is copied directly from Claude into AO3’s editor, so material edited in Google Docs or Microsoft Word will not be flagged. The tag also does not indicate how heavily Claude was used; a red page could reflect a fully AI-generated work or a single sentence pasted for translation or spell-checking. Some flagged authors have edited their works to remove the artifacts, and the tool’s creator acknowledged it was meant to demonstrate capability rather than to accuse individuals.

What context led to this tool appearing now?

Many fanfiction readers and writers have long shared informal "tells" for AI writing, such as certain sentence structures and punctuation choices, but communities have sought more technical solutions. The June 29 post came after years of broad distaste toward tools like Claude and ChatGPT in creative circles, and a growing push to prevent generative AI from reshaping fan spaces. At least one other person claims to have written detection code for "Claude, Deepseek, and some ChatGPT," but that code has not been released or explained. Anthropic did not respond to requests to verify whether the artifact reliably indicates Claude usage.

Why it matters

The skin exposes how fragile technological detection can be and how quickly social enforcement follows. The method finds only a narrow class of copy-paste behavior, yet it has already contributed to public naming and shaming in fandom. That dynamic pressures authors to self-censor, hide tool use, or alter workflows to avoid accidental flagging, and it risks punishing writers who relied on an editor or an intermediary who used Claude without their knowledge. The existence of an AO3 "Created Using Generative AI" tag offers a disclosure path, but backlash reduces the incentive to use it.

What to watch

Look for wider adoption or adaptation of the skin beyond AO3 and for any public release of the other claimed detectors that target Claude, Deepseek, or ChatGPT. Also watch for platform or model-provider responses: Anthropic, Google, and OpenAI were asked about traceable artifacts but did not reply, and any change in model copy-paste behavior or publisher tooling would alter how useful these community-built detectors remain.

The debate is active and practical. For now, the skin is a narrow technical fix that has already escalated social consequences in fandom, and it highlights the broader problem that there is no agreed, reliable way to distinguish human from generated text in most publishing workflows.