Where the Wild Stories Are

For fanfiction writers right now, it can sometimes feel like generative AI is closing in on all sides.

There’s our published work—like last month, when someone scraped the majority of the Archive of Our Own and put the dataset online (responding to criticism of their actions with an image reading, “Ethics: Not needed when you have LLMs”). Or there are our works in progress: the many anecdotes over the past few years about impatient readers feeding WIPs into AI tools to “write” a chapter or “finish” a story. As one anonymous Tumblr asker recently put it, “Maybe if people updated more we wouldn’t turn to ai."

But fic isn’t simply being fed into AI models and tools: in a recursive cycle of sorts, we’re seeing people use those same models and tools to generate stories, too. On AO3, more than 5,000 works have the parent tag “Created Using Generative AI"; perusing the individual tags under that umbrella, you can get a full sense of the range of genAI use, from image and prose creation to grammar checking and translation. “ChatGPT stayed clutch,” one tag reads; another says, “the AI came up with it don’t blame me.”

5,000 works out of the nearly 15 million on the Archive is obviously a drop in the bucket, and on Wattpad and Fanfiction.net, searches for “generative AI” also yield relatively small numbers of results. But given broader AI adoption numbers, it’s highly likely that these tagged and marked stories are just a fraction of current fanworks that are, at least in part, being made with these tools—they’re just the ones posted by people who, for whatever reason, are actually choosing to disclose that fact.

If you’re familiar with ChatGPT’s stylistic quirks, you can spot plenty of probable AI-generated text in the fanfiction wild right now: not “bad writing” by any means, but often notably generic, with a kind of smooth, unvaried shape to both sentences and paragraphs. But it would be totally inappropriate to accuse a fic writer of using these tools—you’d be guessing, obviously, but also, there’s also no “rule” against using genAI to create fic. Literally: on AO3, AI-generated works fall under their maximum inclusivity content policies (and practically speaking, “no AI fic” would be totally unenforceable, and would likely lead to an avalanche of false accusations).

If you browse the works of people who are choosing to disclose, you start to see why someone might not be so eager to tag their fic “ChatGPT stayed clutch.” There’s hand wringing and preemptive defensiveness—a lot of “don’t like don’t read” and promises to delete any comments that complain about the AI component. “I’m not a writer,” one person offers by way of explanation. Others cite the fact that there weren’t many fics for a particular ship, or that no one had written the exact story that they wanted to see.

Fandom is a vast and varied space made up of many overlapping communities, and just like the world at large, attitudes about genAI aren’t a monolith. But since these technologies first arrived on the consumer market a few years ago—and especially as we learned the tech industry slurped up fandom creativity en masse for LLMs—the general sentiment within a lot of corners of transformative fandom towards genAI has been loudly negative.

It’s safe to assume that among fans using these tools, some of the preemptive defensiveness comes from seeing those conversations, or from getting blowback on things they already published. But that loud negativity is likely pushing a lot of other people to mask their AI use, filling up certain fandoms and ships with untagged AI-generated content that’s impossible to filter out.

These tools have landed at a complicated time for the fic world. In the past few years, we’ve seen a rapid rise in the number of readers, but people across fandoms are reporting a steady decline in active creators—people making fanworks, or doing “fandom infrastructure” sort of practices like reccing, prompting, and organizing fests. Fic has historically had a high ratio of writers to readers compared to the pro-writing world, but as readership increases and participation decreases, this gap is widening, too. There’s more passive consumption, and less desire to make stuff yourself.

And then there’s what might be described as the professionalization of fanfiction—the pressure many writers feel to “market” their fic. Unlike fanart, fic remains largely non-monetized, but even without money changing hands, social media has fundamentally altered the way many people frame the stories they’re putting out in the world. We’re seeing people deliberately mistagging stories for greater “reach,” or deleting their work if they feel, stats-wise, it has “flopped,” or posting about how AO3’s algorithm didn’t like their story—when there is, of course, no AO3 algorithm.

All of these factors combine to create a sort of petri dish for genAI use. Take the standard fannish urge to explore a character or a ship you’re obsessed with, or to see a story idea play out in fic form. Add the wide availability of these tools, particularly among current and recent students, who, by their own accounts, have already been using them heavily in academic contexts. Top it all off with the desire to make something polished enough to be marketable to an ever-growing pool of readers—and the near-universal frustration that your own work isn’t as good as the thing you envisioned in your head.

Fanfiction has always been an amateur practice—and that means it’s always been messy. We find ourselves downplaying this element when we try to defend fic to sneering outsiders: “It’s not all bad writing,” we insist. “Some of it is as good as any published literature—or better.” Sure, good writing versus bad is a subjective judgment, but fic has always had space for every kind of writing, and that includes the nonsensical plots, the stilted dialogue, the incorrect spelling and punctuation. If you’ve spent any time in fandom, you’ve certainly come across stuff you could barely parse—but you knew that someone wanted to write it, and that they chose to put it out into the world.

Why should I bother to read what you couldn’t be bothered to write?

There’s an oft-repeated phrase when it comes to AI-generated prose: Why should I bother to read what you couldn’t be bothered to write? This sentiment certainly applies to genAI fanworks, but there’s something more in those works’ lack of fannish affect. You could make an argument that AI-generated fanfiction isn’t fanfiction at all—the words aren’t penned by a fan, but rather by a series of statistical probabilities drawn from a giant database. Yes, a fan is doing the prompting (or the chatting, as transcripts from sites like Character.AI are repurposed as fic). But that messiness is gone: the fannish fingerprints for the thing you love have been totally wiped clean from the work.

In a recent essay in The New York Times, novelist Tom McAllister wrote about how genAI is reshaping his students’ work. He describes a prompt in his memoir-writing class where he asks students to write about a pop-culture “obsession,” explaining that, “This assignment invariably produces a few surprising, insightful and energetic pieces from students who were never asked to take their own interests seriously before.” But this year? “I received a 2,000-word submission defining ‘obsession,’ citing the D.S.M.-5 and multiple online sources,” McAllister writes. “All written in the trademark flat, lifeless prose of ChatGPT.”

Writing about a favorite movie or video game for a class is, of course, still an academic assignment, but there’s a clear parallel between this anecdote and the use of genAI in fandom. McAllister’s essay is titled “I Teach Memoir Writing. Don’t Outsource Your Life Story to A.I.” The word “outsource” is a perfect one for fandom AI use—because when people let the bot write the fic, they’re outsourcing their own fannishness.

For all that people moan about how hard writing fic can be—and perhaps that shared rhetoric contributes, at least in part, to the perception that it’s an unpleasant task that should be avoided—writing is the whole point of creating fic. We’re spending time with characters who fascinate us, and we’re making them do whatever we want: kissing, or rising from the dead, or working service jobs where they somehow have genuine sexual interest in a customer. We’re doing it for free because we enjoy it: the experience of creating these works, and the experience of sharing them with others.

If you outsource the act of being a fan to AI, what does that leave you? Fan creators are powerful because they’re deeply participatory media consumers—they don’t passively absorb a work, but grab onto it and reshape it to their will. Large tech and entertainment corporations prefer the passive: they want us sitting there, clicking a button, as stories wash over us like the automatic scroll of a video app. Next, next, next.

In a post that’s been making the rounds recently, the writer Christopher Buecheler described his bafflement when a friend asked why he didn’t feed just his work into AI so it could have the characters react to plot points. One respondent, PT, replied, “I love imagining my characters in various scenarios and all the intricate, complex ways they react to things. That's literally the entire reason they exist, that's the FUN PART.”

Fanfiction is that FUN PART, from start to finish. We get to spend as long as we want thinking about our favorite characters, but we also get to put all that down on the page, with no real rules or restrictions—no length or genre limitations, no publishing-industry pressures, no reader to satisfy but ourselves.

It’s been heartening to see the wide rejection of generative AI in transformative fandom spaces—it’s an indicator of just how much the broader community values real human creativity, which isn’t always easy to see when you’re toiling away in some small corner of fandom. Hopefully the fans who think they aren’t writers, or that they couldn’t write something “better” than ChatGPT, come to learn that there are no barriers to being “a fanfiction writer”—and anything created by a fan, with all those fannish fingerprints, will always be superior to whatever’s spit out by the machine.

Elizabeth Minkel is journalist, editor, and consultant who focuses on digital technologies and fan culture. She’s written for WIRED, Atlas Obscura, The Guardian, The New Yorker, the New Statesman, and many other publications. She’s the host and editor of Fansplaining, a podcast and publication about fan culture, and she’s the co-curator of the fandom newsletter “The Rec Center,” which was a finalist for the Hugo Award. She has a master's degree in the digital humanities from University College London. She lives in Brooklyn with her cat, Orlando.