How Creative Is AI Writing? Generative and Collaborative AI in Japanese Fiction

Abstract

This study examines readers’ perception of quality and originality in AI-generated and human–AI collaborated Japanese literary fiction. Fifty-eight participants with backgrounds in creative writing ranked the endings of 25 stories across four conditions. In all cases, the first half of each story was authored by professional or amateur writers; the second half varied by condition: original human-authored text, zero-shot AI generation, AI generation with detailed prompt specification and human–AI collaboration combining multiple-shot prompting with human editing. Focus group sessions followed to explore the criteria underlying the ranking provided by participants. Human–AI collaboration received the highest mean quality rankings, followed by original human writing, detailed prompt specification and zero-shot generation. While the differences between conditions were small relative to within-condition variance, the contrast between human–AI collaboration and zero-shot generation was statistically significant. In addition to that, collaboration was rated highest in perceived quality but lowest in originality, a dissociation that previous research had not clearly identified: AI-assisted editing produces polished, coherent prose that readers find accessible, but at the cost of the stylistic expression that defines literary fiction. These findings suggest that the contribution of generative AI to creative writing is better characterised as a reduction of variance than a tangible elevation of quality.