A study from the University of Washington and the University of Colorado Boulder analyzing the WildChat dataset reveals that 34% of user interactions with ChatGPT are dedicated to fiction generation, often driven by a small cohort of “super users” who engage in repetitive, niche storytelling. Researchers identified these patterns by examining over half a million anonymized conversations, noting that fan fiction and erotica represent the most popular genres for these high-volume users.
How Researchers Identified "Infinite Story Demanders"
In the paper titled “AI Fiction in the Wild,” researchers utilized the WildChat dataset, a collection of over 573,453 English-language conversations between humans and ChatGPT. The team found that fiction generation—ranging from original novels to roleplay—accounted for 195,271 of the total interactions.
The study highlights a specific trend of “infinite story demanders,” users who repeatedly prompt the AI for the same highly specific narratives over extended periods. One notable outlier identified in the study frequently requested stories involving characters from the 2017 visual novel Doki Doki Literature Club. These prompts often involved the same characters in specific, repetitive scenarios, with the user providing long dialogue segments that the model was instructed to continue.
The Role of Heavy Users in AI Fiction
The data indicates that fiction generation on the platform is not evenly distributed. According to the researchers, the top 2% of users interested in fiction were responsible for 80% of the total fiction-related conversations. This concentration suggests that approximately 200 individuals generated 150,000 fiction prompts within the dataset.

Researchers categorized these prolific users into two primary groups:
- Infinite story demanders: Users who request the same specific story variations for months at a time.
- Story cyclers: Users who generate multiple versions of a single narrative arc before transitioning to a new topic.
What Genres Dominate AI-Generated Fiction?
The analysis of the WildChat dataset shows that fan fiction (49%) and erotica (29%) are the most common categories for AI-generated stories. The researchers noted that the high volume of these genres reflects the platform’s utility for users seeking instant responses tailored to niche preferences.

The content frequently triggers safety filters; the study reports that 10% of these fiction conversations were flagged as explicit, while 17% were identified as toxic by moderation tools such as OpenAI Moderation or Detoxify. Common sources for these fan-fiction prompts included popular media franchises such as Naruto, League of Legends, Freedom Planet, and Doki Doki Literature Club.
Key Takeaways
- High Concentration: A small group of "super users" accounts for the vast majority of fiction-writing prompts on ChatGPT.
- Genre Popularity: Fan fiction and erotica are the leading categories, making up nearly 80% of identified fiction conversations.
- Repetitive Behavior: Prolific users demonstrate a high degree of repetition, with the top 10 most active users repeating prompts 85% of the time.
- Moderation Trends: A significant portion of these requests involve content that triggers automated toxicity or explicit content flags.
The findings underscore how specialized user groups utilize generative AI as a tool for personal creative expression, often focusing on narrow, iterative storylines that deviate from standard conversational use.