AI-Generated Characters Are Too Neat, Study Finds
The study suggests that current AI writing tools, regardless of scale, may produce stories that feel predictable because they avoid the ambiguity that makes characters memorable.
Reporting from 1 sources: GIGAZINE.
Researchers at the University of North Carolina at Chapel Hill analyzed character variety in stories written by large language models versus humans. Using a framework called CASPER, they found AI characters tend to be archetypal and resolve neatly, while human writers leave ambiguity and contradiction. Larger models did not produce more complex characters.
Researchers at the University of North Carolina at Chapel Hill developed an automated framework called CASPER to evaluate character traits in AI-generated fiction across eight axes, including stylization, consistency, and closure. The study, presented at the ACL 2026 conference, compared thousands of stories from large language models with human-written narratives.
Lead author Annelise Bray said AI models take the 'safe route' by wrapping up characters neatly, whereas human writers sometimes leave questions unanswered. The paper found that larger models did not significantly improve at producing ambiguous or contradictory characters, a trait the researchers argue is key to stories that linger in a reader's mind.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.