One ChatGPT User Generated Nearly 28000 Fanfiction Chats and It Gets Even More Bizarre From There

AI is turning fiction into an endless, private, personalized loop.

Somewhere on the internet, one anonymous person spent months asking ChatGPT to thousands of variations of the same fanfiction. Normally, we’d chalk this up to the “weird corners of the internet” aisle. But now, it’s part of a remarkable new study.

According to a new study, the most prolific fiction user in a large public chatbot dataset generated a staggering 27,869 conversations around similar story prompts, many involving Doki Doki Literature Club! characters and a weirdly recurring pregnancy plotline. The researchers called this user “the clearest example of an infinite story demander” — someone who keeps feeding the machine nearly the same premise, waiting for a slightly different version to come back.

In their new paper aptly titled AI Fiction in the Wild, researchers from the University of Washington and the University of Colorado Boulder analyzed more than 573,000 anonymized English-language ChatGPT conversations from the WildChat dataset, a public collection of opt-in chatbot logs gathered through Hugging Face. They found that more than a third involved prompts for generating fiction: stories, scripts, roleplay, worldbuilding, fanfiction, erotica, or some unruly blend of all of the above.

The Infinite Story Machine

Doki Doki Literature Club! is one of the strangest computer games out there. It’s a 2017 visual novel that starts out looking like a cute high-school dating game, but then swerves hard into psychological horror. It breaks the fourth wall in unusually aggressive ways, deleting files on the users’ computer and manipulating its own storyline in bizarre ways. Apparently, the game has developed quite the fandom.

The anonymous fan seems to have used ChatGPT like a never-ending narrative vending machine. Across thousands of prompts, they returned to the same basic setup, the same characters, and the same crisis, changing only small details each time. Then they did it again. And again.

The plot had the fever-dream quality of someone who knew exactly what they wanted and absolutely did not want to explain it to a human being. In the school clubroom, Natsuki is heavily pregnant with a daughter named Sakura. Her water breaks, so her friends panic. Someone needs to call an ambulance. Natsuki screams, grips Yuri’s hand, and announces, in so many words, that the baby is coming now. The user often cut the scene off mid-crisis, leaving ChatGPT to pick up the sentence like an improv partner trapped in a maternity ward.

The model dutifully filled in the blanks, including the breathing instructions, the emergency responders, and, in one version, the paramedic’s happy-ending announcement: “Congratulations, Natsuki. It’s a beautiful baby girl.”

In this case, the same digital curtain rose thousands of times on the same scene — Natsuki, Sakura, the clubroom, the ambulance — as if Groundhog Day had been rewritten by a fan forum with a very specific maternity subplot.

×

Get smarter every day...

Stay ahead with ZME Science and subscribe.

Daily Newsletter
The science you need to know, every weekday.

Weekly Newsletter
A week in science, all in one place. Sends every Sunday.
No spam, ever. Unsubscribe anytime. Review our Privacy Policy.

Thank you! One more thing...

Please check your inbox and confirm your subscription.

The researchers argue that this behavior may not necessarily reflect dissatisfaction with the output.

“Some of this behavior may be attributed to users’ dissatisfaction—the story is not quite right, at least not yet,” they write in the paper. “But we argue that in many cases it is fueled by the satisfaction of reading endless permutations of the same story, none of which end the same way twice.”

The paper calls these users “infinite story demanders.” Others are “story cyclers,” who obsess over one setup for a while, then move to another. One user repeatedly prompted time-travel melodramas about a fictional news editor trying to undo her parents’ forced marriage. Another kept returning to familiar characters, tropes and fake-news-style scenarios.

The prose was often cliché, and ChatGPT wasn’t exactly producing Pulitzer-worthy materia. But that’s beside the point. Many users appeared to want recognizable formulas and happy endings, and on that, the machine delivered in spades.

Fanfiction Without the Fandom

The researchers classified 195,271 conversations (34% of the English-language dataset) as fiction. Of those fiction conversations, 95,450 were fanfiction and 52,231 contained sexually explicit material.

The numbers were lopsided. The top 2 percent of fiction-generating users produced more than 80 percent of the fiction conversations. Among the ten most prolific users, prompt repetition reached 85 percent.

Fanfiction is fiction written by fans using characters or worlds that already exist — Harry Potter, Naruto, Star Wars, Doki Doki Literature Club!, anything with a devoted audience. Traditionally, a fan writes a chapter, posts it on a site like Archive of Our Own or Wattpad, and other fans read it, comment on it, argue about it, praise it, ask for more, or suggest where the story should go next. The writer may update the story days, weeks, or months later. The reader waits, like everyone else.

But some AI users no longer need another fan to write the next chapter. They can ask ChatGPT for exactly the version they want: this character, that romance, this crisis, that ending, more dialogue, less tragedy, happy ending, try again. Instead of joining a shared fan community, the reader can sit alone with the machine and generate a custom story on demand — not the story someone else imagined, but the one they ordered.

The authors call this emerging figure the “solipsistic reader-writer”: someone who generates and consumes fiction in a closed loop with a machine, without another human storyteller or, often, other readers for that matter.

That does not mean all AI fiction is lonely or empty. Some people may use chatbots to brainstorm novels, run tabletop games, amuse friends, or write stories they later share. But the WildChat data points to a different literary pattern, where new fiction is generated as instant gratification.

Not Just Slop

It is tempting to laugh this off as disgusting AI slop. Futurism, where we first saw this story, called this prompter the “most degenerate ChatGPT user.” The researchers do not agree.

They argue that AI fiction belongs to a longer history of self-publishing, fanfiction, interactive storytelling and personalized entertainment. The main and most obvious difference is speed. A fanfic author may take weeks to update a story. A chatbot takes seconds.

Tech companies have noticed. Google’s Gemini Storybook now creates personalized illustrated stories with narration from a prompt. Character.AI, which has positioned itself as an AI entertainment company, has drawn tens of millions of users to character-based chatbot interactions.

RelatedPosts

Google’s AI is now suggesting new football tactics
This Ancient Roman Scroll Was Burned to A Crisp 2,000 Years Ago. Now, Researchers Are Reading What’s Inside With AI
AI Tool Reveals Signs Of Consciousness In Comatose Patients Days Before Doctors Can Detect It
OpenAI Model Cracked an 80-Year-Old Math Problem and Mathematicians Are Stunned

The WildChat dataset is not representative of all ChatGPT users. Still, it offers a glimpse into what some people actually do when handed a machine that can endlessly improvise.

And what some people do, apparently, is ask for the same story 27,869 times.