Readers turn to AI chatbots for personalised fiction and role-play
A University of Washington analysis found more than one-third of prompts in a large WildChat dataset involved fiction generation, though researchers cautioned it was not representative of all chatbot users.
Readers are using AI chatbots to create personalised fiction, fan fiction, erotica and role-play, shifting attention from how authors use the technology to how audiences consume it.
A University of Washington-led analysis of more than 500,000 anonymised prompts from the public WildChat dataset found that more than one-third involved fiction generation. The category included prose, poetry, fan fiction, erotica and role-play. After accounting for heavy users, researchers estimated that about 7 per cent of chatbot users engage in fiction generation.
The findings, reported by WIRED, suggest some users are attracted to the speed of generating and revising stories, direct control over plot and characters, and interaction without judgement. Models can also support highly specific fantasies that may be difficult to find in published fiction or online pornography.
Petra Ferraz de Novaes, a software engineer from Brazil, uses SillyTavern and open-source models to build fantasy stories and private role-plays. She said the unpredictability of AI-generated responses was part of the appeal and that the experience had encouraged her to read other people’s AI-generated novels.
The researchers cautioned that WildChat is self-selected and cannot be treated as representative of all chatbot users. The estimate also differs from a 2025 OpenAI and National Bureau of Economic Research study, which found that 1.4 per cent of ChatGPT prompts were fiction-related; the measures may differ because that figure excludes role-play and may exclude erotica.
Current models perform better with familiar genres and established tropes than with ornate prose, complex characters or coherent book-length narratives. The University of Washington researchers said those limitations could change as models improve, while also raising questions about whether AI-generated entertainment could expand alongside a smaller market for experimental human-authored fiction.
