CASPER in the Machine: Insights into Character Variety in LLM-Generated Stories

Researchers analyzed how characters in LLM-generated stories differ from human-written narratives across eight dimensions including stylization and wholeness. The study reveals meaningful differences in character complexity and variety between AI-generated and human fiction, raising questions about the depth of LLM storytelling capabilities.

This academic research addresses a critical gap in understanding how large language models handle narrative complexity, specifically through character development—a cornerstone of compelling fiction. The study applies narratology frameworks to systematically compare LLM-generated stories with human-written works, moving beyond surface-level metrics to examine nuanced character dimensions like stylization and psychological wholeness.

The research emerges from growing concerns about AI-generated content quality as LLMs become embedded in creative industries. While previous analyses focused on plot coherence or grammatical correctness, examining character depth provides insight into whether LLMs truly grasp the psychological and emotional nuances required for engaging storytelling. This matters because character consistency and variety are what distinguish memorable fiction from formulaic content.

For the content creation and entertainment industries, these findings have practical implications. Companies integrating LLMs into creative workflows need to understand where these models excel and where human oversight remains essential. Publishers, game developers, and streaming platforms using AI-assisted content generation must recognize character-related limitations to maintain narrative quality standards. The analysis of whether LLMs generate diverse character archetypes also informs content moderation and representation concerns.

Looking forward, these findings should drive further investigation into whether current LLM architectures can be enhanced to produce more psychologically complex characters, or if narrative depth fundamentally requires human creativity. The research provides a foundation for developing evaluation metrics that extend beyond technical performance to assess creative quality—a necessary step as AI-generated content becomes commercially significant.

• →LLM-generated stories exhibit measurable differences in character complexity compared to human-written fiction across eight distinct narrative dimensions
• →Character variety and stylization analysis reveals limitations in how LLMs develop psychologically coherent fictional personas
• →The research demonstrates the need for human-centered evaluation frameworks beyond basic metrics to assess creative AI output quality
• →Understanding character depth gaps in AI-generated narratives is critical for industries integrating LLMs into content creation workflows
• →This study provides methodological foundations for developing more sophisticated evaluation tools for creative AI applications