Researchers analyzed over 500,000 anonymized ChatGPT conversations and discovered that more than one third involve fiction generation, including original stories, roleplay, fanfiction, and erotica. The study reveals that AI-generated fiction is dominated by power users who repeatedly request variations of similar narratives, creating what researchers term 'infinite story demanders.'

These users show strong preferences for fanfiction and erotica, gravitating toward generic forms, repetition, and niche story element combinations. The findings suggest a fundamental shift in how people interact with narrative content, moving from passive consumption to active co-creation with AI systems.

The research introduces two provocative concepts: the 'solipsistic reader-writer' who generates and consumes fiction within a closed conversational loop, and the idea that LLMs enable new forms of interactive storytelling. Unlike traditional media consumption, users engage in continuous revision and permutation of narratives.

This work situates AI fiction generation within broader cultural trends like self-publishing and on-demand content. The structural similarities to personalized, repetitive cultural forms suggest AI tools are creating new entertainment paradigms that prioritize immediate, customizable narrative experiences over conventional publishing models.