Updated
Updated · AUTOMATON WEST · Jul 1
Study of 573,000 ChatGPT Logs Finds Top 2% Drove 80% of Fiction Prompts
Updated
Updated · AUTOMATON WEST · Jul 1

Study of 573,000 ChatGPT Logs Finds Top 2% Drove 80% of Fiction Prompts

1 articles · Updated · AUTOMATON WEST · Jul 1

Summary

  • 195,000 of roughly 573,000 English ChatGPT conversations were classified as fiction, and researchers highlighted one outlier who generated thousands of Doki Doki Literature Club! pregnancy fanfics over months.
  • 52,000 fiction chats contained sexually explicit material—nearly 30%—while the study found fiction use was heavily concentrated among a small set of repeat users.
  • Top 2% of fiction-generating users, about 200 people out of an estimated 10,000, accounted for more than 80% of fiction logs, with 69% of their prompts being repetitive reruns or refinements.
  • Doki Doki Literature Club! led franchise mentions at 22,381, far ahead of Freedom Planet at 5,204, League of Legends at 4,514 and Naruto at 4,342.
  • The paper used the WildChat dataset—consented, anonymous chats from a free Hugging Face-hosted bot between April 2023 and May 2024—and said it offers a useful but nonrepresentative view of ChatGPT fiction use.

Insights

Should app stores or AI creators be held responsible for policing users' AI-generated fanfiction?
Is AI turning fiction into a private obsession rather than a shared cultural experience?
As AI writes more of our stories, who will truly own the copyright to human creativity?