AI Fiction in the Wild

More than one third of anonymized ChatGPT conversations involve fiction generation, according to a new study that analyzed over 500,000 English-language user interactions, revealing a landscape dominated by power users and marked by a strong appetite for fanfiction and erotica [1][2]. The preprint, titled "AI Fiction in the Wild" and posted to arXiv on 22 June 2026, draws on a dataset of anonymized ChatGPT-user conversations to map how readers are using large language models to produce fiction [1][2]. The authors, led by Melanie Walsh, find that fiction generation is not a fringe activity but a central use case, with more than one third of the sampled conversations containing some form of narrative creation [1][2]. The study identifies a cohort of "power users" who dominate this output, including a group the researchers call "infinite story demanders" — users who repeatedly request and revise variations of the same or similar narratives over extended periods [2]. The analysis shows users gravitate toward fanfiction and erotica, and are broadly drawn to generic forms, repetition, immediacy, and niche combinations of story elements [1][2]. The researchers argue that these patterns share structural affinities with on-demand, personalized, and repetitive cultural forms already present in self-publishing, fanfiction, and pornography [2]. The paper introduces the concept of a "solipsistic reader-writer," describing a user who both generates and consumes fiction within a closed conversational loop, interacting with a machine rather than a human other [1][2]. The findings arrive amid growing public discussion about the quality and volume of AI-generated content. The term "AI slop" — defined as digital content made with generative AI that is perceived as lacking effort, quality, or meaning and produced in high volume — was selected as the 2025 Word of the Year by both Merriam-Webster and the American Dialect Society [3]. Jonathan Gilmore, a philosophy professor at the City University of New York, has described such material as having an "incredibly banal, realistic style" that is easy for viewers to process [3]. The study's documentation of repetitive, on-demand story generation echoes these concerns about content prioritizing speed and quantity over substance [2][3]. Broader ethical questions around AI-generated media also frame the research. The ethics of artificial intelligence encompasses algorithmic biases, accountability, transparency, and the risks of AI-enabled misinformation, all of which intersect with the production and distribution of synthetic fiction [6]. While the arXiv paper does not directly address these ethical dimensions, its portrait of users locked into personalized narrative loops raises questions about how such systems might shape reader expectations and the broader storytelling ecosystem [2][6].