The Heated Rivalry AO3 AI Controversy, Explained

On June 29, 2026, an anonymous account on X posted a 25-page document titled “Fandom Has a Hidden Generative AI Problem.” It described a way to spot fan fiction that had been copied straight out of Claude, Anthropic’s chatbot, and pasted into the rich-text editor on Archive of Our Own (AO3). Text moved that way could carry a Claude-specific chunk of HTML formatting that readers never see on the page but that stays in the work’s code.

The document listed 38 works where that formatting turned up. Many came from the fandom around Heated Rivalry, the TV series adapted from Rachel Reid’s Game Changers novels, whose fan fiction exploded after the show aired at the end of 2025. Fansplaining reported that the fandom grew from around 300 works on AO3 in November 2025 to over 40,000 by July 2026. Works from other fandoms, including The Pitt, Pluribus, Off-Campus and Stranger Things, were on the list too.

The document also shipped an AO3 site skin, a piece of custom styling a reader can install, that highlights the hidden Claude formatting on any work they open.

The list landed on a fandom that was already tense. According to Fansplaining’s July 2026 reporting, popular writers had spent the spring dealing with demands for faster updates and accusations of undisclosed AI use, some based partly on how quickly they posted. Some authors had denied using AI and received waves of sympathy. When their works appeared on the list, readers who had defended them felt betrayed; readers who had suspected them felt vindicated.

The document’s authors said they wanted transparency, so readers could choose what to read, and explicitly condemned harassment of the people it identified. That did not hold. Fansplaining and IBTimes UK both reported that more works were deleted or locked in the week after the release, and that at least one writer turned off comments after receiving death threats. Several authors added notes explaining how they had used AI, for example for proofreading, formatting or in place of a human beta reader.

We are not naming any writers here. Whatever you think of undisclosed AI use, a public list of names plus a detection tool is a recipe for dogpiling, and the people on the receiving end are real.

The hidden formatting is far stronger evidence than a vibe or a detector score, because it is a physical trace rather than a guess about style. If a work contains it, some text in that work very likely went directly from Claude into AO3’s editor.

But its reach is narrow, as both Fansplaining and IBTimes UK spelled out. It says nothing about text pasted into another document first, text where the formatting was removed, or text from ChatGPT, Gemini or any other model. It cannot tell you how much of a story came from Claude or what Claude was used for: drafting a chapter and fixing commas leave the same mark. Some flagged works carried two instances of the wrapper; others carried hundreds.

The document’s own authors said the absence of the code is not evidence that AI was not used. And by publishing the method, they also showed exactly how to avoid it, which means the trace will probably get rarer over time. Treat it as proof of one specific workflow, not as an AI detector.

AO3 does not ban AI-generated works. In a May 2023 post, the Organization for Transformative Works, the nonprofit that runs the archive, said nothing in its Terms of Service prohibits fanworks that are fully or partly generated with AI tools, as long as they otherwise qualify as fanworks, and that its goal of maximum inclusivity covers them. It added that mass-posting AI works can still break its anti-spam rules.

Fansplaining quotes a 2024 AO3 Terms of Service update guide on why a ban is off the table: humans, including the volunteer Policy & Abuse team, cannot reliably detect AI use, and AI-detection programs are “extremely unreliable and prone to false positives.” IBTimes UK reported that the OTW restated that position in June 2026. Disclosure is optional, through tags such as “Created Using Generative AI”, which Fansplaining counted at about 10,000 works out of more than 17 million on the archive.

One more detail matters for anyone reading accusations in their comments. The OTW’s living guide to spam on AO3 lists “AI Use Accusation Spam”: guest comments, sent by bots, that accuse writers of using AI, sometimes naming a detector or claiming to have seen prompts removed. Not every accusation comes from a person.

For readers, the honest position is uncomfortable: you cannot reliably tell. A detector score is not proof, which we cover in more depth in can AI detectors prove a novel was AI-written?, and a missing trace is not proof either. What you can control is where you spend attention: follow writers whose process you trust, and use tags and author notes as the signal they are meant to be.

For writers, the lesson is that disclosure is cheaper than discovery. IBTimes UK cited a study of 157 fan-fiction readers and writers in which 86% said authors should disclose AI involvement, and 58% said they would feel deceived to learn later that a story was AI-written. The writers hurt most in this episode were not necessarily the ones who used AI the most. They were the ones whose stated position did not match what readers later found.

Disclosure also needs to be specific to be useful. “AI-assisted” covers everything from a grammar checker to a full draft. A line such as “I write every scene myself; I used an AI tool for proofreading” tells readers far more than a tag alone.

ShortFling is not a fan fiction site. Our fiction library is for original stories, and our creator agreement requires that work published through the editor is original and that every character is an adult. The editor includes an AI assistant, so the transparency question applies to our writers too.

Our view is the one this controversy points toward: if AI helped shape a story, say how in your own words, and keep your drafts so you can show your process if anyone asks. Publishing on ShortFling submits a story for review before it goes public, but review is not an AI verdict, and we do not offer AI detection or certification. For other places to publish adult fiction, see our guide on where to publish erotica online.