A study of more than 500,000 ChatGPT prompts suggests that fiction writing—including prose, fan fiction, erotica and role-play—accounts for over a third of chatbot interactions, with roughly 7% of users regularly engaging in it once heavy "power users" are accounted for. The research, led by University of Washington academic Melanie Walsh, pushes back against the narrative dominating the publishing world, where recent scandals over AI-assisted authors have implied that readers universally reject machine-generated fiction. Instead, the findings suggest a dedicated subset of users happily create and consume AI-generated stories, valuing the technology precisely because it produces unpredictable, personalised narratives.

The study drew on the public WildChat dataset, though its authors acknowledge limitations: prompts are anonymised with no demographic data, and users had to consent to making them public, so the sample may not represent all ChatGPT users. A separate 2025 study by OpenAI and the National Bureau of Economic Research put fiction-related prompts at just 1.4%, though that figure excludes role-play and is unclear on erotica. Users such as Brazilian software engineer Petra Ferraz de Novaes, who builds elaborate fantasy worlds using the open-source tool SillyTavern, describe the appeal as the AI's unpredictability and its capacity for near-endless, self-directed storytelling.