A recent study presented at Purdue University’s MFS Cultural AI Conference has pulled back the curtain on a peculiar, often repetitive corner of the internet. The paper, titled AI Fiction in the Wild, analyzed a massive dataset of 500,000 anonymized English-language ChatGPT conversations. While the researchers set out to map the landscape of AI-assisted creativity, they stumbled upon something far more complex: a world of "infinite story demanders" and hyper-fixated users who are fundamentally altering the way fiction is produced and consumed in the digital age.
The data, sourced from the WildChat interface hosted on Hugging Face, offers a rare, unfiltered look at how humanity interacts with Large Language Models (LLMs) when the guardrails of public social media are removed. What the researchers discovered wasn’t just a surge in creative writing, but a psychological shift in how we engage with the stories we love.
The Taxonomy of AI-Prompting: Cyclers vs. Demanders
The study’s findings indicate that fiction generation is a primary use case for generative AI, with more than a third of all analyzed conversations involving some form of narrative creation. This ranges from scripts and world-building to roleplay and erotica. However, the researchers identified two distinct archetypes among these users:
- Story Cyclers: These individuals operate with a degree of narrative variety. They request iterations of a story for a set period, experiment with themes, and then pivot to new topics or characters once their interest wanes or the narrative arc reaches a natural, if AI-assisted, conclusion.
- Infinite Story Demanders: This group represents a more intense, repetitive form of engagement. These users demand the same story—or nearly identical variations of it—over and over again for weeks or even months.
The researchers noted that the "infinite" nature of these prompts suggests that for many, AI is not a tool for creating something new, but rather a mechanism for experiencing a specific, comforting, or obsessive feedback loop.
The Case of the DDLC Obsession: A Chronology of Data
Perhaps the most startling revelation from the AI Fiction in the Wild report was the discovery of a single, highly prolific user whose fixation defied typical usage patterns. Over a period of several months, this user generated thousands of variations on a singular narrative arc: a character from the visual novel Doki Doki Literature Club (specifically Natsuki) giving birth to a daughter named Sakura.
This behavior highlights a striking statistical imbalance: while the dataset contains hundreds of thousands of users, the research found that just two percent of users within the fiction-generating subset were responsible for more than 80% of the total conversations.
The inclusion of Doki Doki Literature Club in these prompts is particularly ironic for those familiar with the game. A psychological horror experience that deconstructs the tropes of dating sims, DDLC is famous for its meta-narrative subversion. Using an AI to force the game’s characters into a repetitive, domestic, and non-canonical loop of pregnancy narratives creates a layer of irony that is as disturbing as it is fascinating to data scientists. Other frequently requested universes included League of Legends, Freedom Planet, and Naruto, suggesting that the most popular fandoms for AI exploitation are those with strong character aesthetics and established, recognizable tropes.
Supporting Data: Comparing WildChat to AO3
To understand the cultural impact of this trend, researchers compared the WildChat data to the long-standing fan-fiction repository Archive of Our Own (AO3). The contrast is stark.
On platforms like AO3, fanfiction is a communal activity. Writers publish chapters, receive feedback, engage in "shipping" wars, and participate in a wider ecosystem of literary critique and celebration. When a fan writer moves from Harry Potter to My Hero Academia, they often bring their audience with them. It is a social, collaborative endeavor that relies on human empathy and shared enthusiasm.
In contrast, the AI-generated fiction found in the WildChat dataset is intensely solipsistic. The AI does not offer critique, nor does it form a community. It simply obeys. This creates a "masturbatory" loop of content generation where the user is the sole audience and the sole creator, insulated from the external validation—and the external challenge—of a real-world community. While the data remains anonymized, the sheer volume of repetitive prompts suggests that for a significant minority of the population, AI is being used as a private, bespoke mirror for their own specific, perhaps even shameful, fetishes or emotional needs.
Implications for AI Development and Society
The AI Fiction in the Wild study has profound implications for the future of AI development. If a significant portion of user engagement is driven by individuals who want to trap an AI in an infinite, repetitive loop, it suggests that current LLMs are being used as "emotional crutches" or "wish-fulfillment engines" rather than creative assistants.
The Problem of "Model Drift" and Feedback Loops
When users force an AI to generate the same story thousands of times, they are essentially training the model—or at least interacting with its alignment—in a way that reinforces specific tropes. If these "infinite demanders" continue to dominate the usage statistics, developers may find that their models are increasingly optimized to handle repetitive, narrow-focus requests rather than broad, creative problem-solving.
The Ethics of Privacy in Research
The data used in this study, while anonymized, raises important questions about the nature of the "public" internet. Because users accessed the WildChat interface on Hugging Face with the understanding that their data could be used for research, they inadvertently provided a window into their private thoughts. This creates a moral gray area: at what point does the scientific value of analyzing human behavior override the individual’s expectation of privacy, even when that privacy was technically waived?
The Death of "Community" in Fandom
If the future of fanfiction moves away from open repositories like AO3 and toward private, AI-driven silos, we may see a decline in the social health of fandoms. Fanfiction has historically been a gateway for young writers to develop their voices and learn the mechanics of storytelling. If that practice is replaced by an AI that never corrects, never critiques, and never tires, we risk creating a generation of writers who are incapable of engaging with a real audience.
Conclusion: A Mirror, Not a Window
The AI Fiction in the Wild report is not just a collection of odd statistics; it is a warning. As we continue to integrate generative AI into our daily lives, the distinction between "creative tool" and "private obsession" is becoming increasingly blurred.
The study confirms that while technology can mimic the structure of human storytelling, it cannot replicate the spirit of it. Human fiction is meant to be shared, debated, and evolved. The "infinite story" generated by an AI in a vacuum, however, is a static, repeating echo—a digital artifact of a user talking to themselves, over and over, in the dark. As the researchers at Purdue University have shown, when you give people a machine that will say anything they want, what they often want is to be trapped in a story that never changes.
