The publishing industry has long debated AI's impact on professional authors, but analysis of real-world ChatGPT usage reveals an unexpected development: readers themselves are becoming prolific fiction generators. The WildChat dataset—over 500,000 anonymized English-language conversations—shows that 34% of interactions involve fiction generation. More striking: this content is dominated by power users who repeatedly iterate on the same narratives, sometimes for months, creating what researchers term "infinite story demanders".

Story cyclers

iterate on the same narrative for a period of time before switching to another story or topic. One documented user generated over 900 conversations across eleven months, transitioning between eclectic genres like time-travel narratives, alternative Hollywood histories featuring Matthew Broderick, and fake news articles.

Approximately half of WildChat fiction conversations were classified as fanfiction, with over a quarter labeled sexually explicit. These genres lend themselves particularly well to LLM affordances: they rely on predictability and formulaic plots, enable variation within established frameworks, and cater to niche combinations of elements.

Fanfiction fundamentally rests on opening canonical source material to alternative configurations. LLMs excel at this, rendering exploration tasks redundant by letting readers specify the exact story parameters they want in natural language. Furthermore, because LLMs are trained on millions of cultural objects, they can produce writing specific to diverse fictional universes—accurately deploying precise diegetic details while remixing them into new configurations.

As models become more capable, accessible, and loose with content restrictions, consumption may accelerate. For AI-generated fiction, production requires minimal human labor at the point of generation. While economic research suggests LLMs are driving a large influx of digitally self-published books, many of these works currently suffer from low quality and minimal reader engagement.

Furthermore, without shared texts that communities can discuss, hyper-customized fiction risks atomization into personalized narratives of total social irrelevance.

One documented case illustrates the immediate real-world tensions: a fanfiction author discovered a devoted reader was inputting her ongoing chapters into ChatGPT to generate the next installment rather than waiting for updates. "It hurt my chest," the author wrote. "The AI feels so insulting and violating".

As critic Anna Kornbluh argues, immediacy may be considered "a master category for making sense of twenty-first-century cultural production". But as scholar Mark McGurl warns, an obvious catering to the reader as a customer who is always right risks defeating the purpose of reading fiction entirely, maintaining that "our interest in fiction is in part an interest in encountering different degrees of... otherness".

Ted Chiang's critique of AI art highlights that inputting a prompt requires making very few choices, whereas true art requires making choices at every scale. However, the WildChat data suggests that for certain consumer genres like fanfiction, readers may primarily care about holding the "large-scale choices"—like character relationships and personal configurations—constant, while letting the model compute the small-scale prose execution.

Market disruption scenarios: Will AI-generated fiction disrupt traditional publishing? Current evidence highlights severe technical and aesthetic limitations. A major bottleneck is length: the average ChatGPT response length in the dataset is 385 words, failing to replicate the immersive, long-form experience of traditional fanfiction (averaging 20,000 words) or novels (averaging 197,000 words). For now, models struggle to generate coherent long-form narratives.

The sociality question: The most significant concern raised by researchers is the potential loss of fiction's social function. Fiction historically operates as a node in a broader network of exchange and community—through book clubs, classrooms, tabletop groups, and online platforms like AO3, where enthusiasts love the community as much as the stories themselves. AI-generated fiction, hyper-customized to the individual, may undermine the production of these shared, communal texts.

Shifting landscapes: While mainstream book publishers insist that AI tools will remain more transformative for internal back-end tasks like evaluation rather than consumer content generation, tech companies and disruptors are actively banking on personalized storytelling. Google debuted its Gemini Storybook narrative tool designed to create short illustrated books based on prompts, while the self-publishing company Inkitt plans to leverage AI to hyper-customize stories for readers.

The WildChat analysis reveals patterns that demand serious attention from publishers, authors, and researchers. Three key implications emerge:

• First, AI-generated fiction represents real-world demand for immediate, customizable, and niche narrative permutations that traditional platforms do not provide at the same rate.

• Second, the dynamic between authorial autonomy and reader preference is fundamentally altering. As AI enables individuals to bypass the traditional exchange with a human author, reading experiences risk collapsing into isolated personal fantasies.

• Third, the future of contemporary storytelling will be shaped by how society navigates this loss of sociality. Turning toward this development allows researchers to better understand the shifting creative terrain, whether future literature evolves alongside AI or develops in deliberate defiance of it.