Neat! The shape of human (left) and AI (the rest) fiction.

For those of us who enjoy reading and writing fiction, AI-generated fiction often feels… off. For me, it feels flat and uninspired, even if I can’t always pinpoint why.

By analyzing over 60,000 stories, a new preprint now gives us some clues:

• AI overexplains the story themes. It’s more explicit and moralistic.

• Human authors are more likely to ‘subvert linearity’, as in play with non-chronological plots and incorporate more subplots.

• AI overwrites bodily senses for emotions. It rarely states an emotion, but will overwrite: “a tightening chest and sweaty palms” instead of “felt afraid“.

• Human authors are more likely to ‘engage the outer world’, as in reference other texts, or brands, or authors, or songs…

• AI narratives are less diverse. Human writers use more dialogue, more distinct settings, more subplots, and more morally gray characters. AI resorts to the default (shocker, I know).

(preprint: arxiv.org/abs/2604.03136)