Do AI stories explain their themes more than human stories do?

When StoryScope reduces its 304 extracted narrative features to a compact set of about 30, a coherent contrast emerges between machine and human storytelling. AI stories over-explain their themes — they spell out the moral or meaning rather than leaving it to be inferred — and favor tidy, single-track plots with clean escalation and resolution. Human stories, by comparison, frame their protagonists' choices as more morally ambiguous and exhibit greater temporal complexity: flashbacks, nonlinear structure, discontinuities in chronology. The divergence is at the level of narrative decisions, not prose.

This pattern recurs and even fractures by model. Per-model fingerprints show distinct defaults — Claude produces notably flat event escalation, GPT over-indexes on dream sequences, Gemini defaults to external character description — enabling 68.4% macro-F1 six-way authorship attribution. But the human-vs-AI contrast sits above these idiosyncrasies: across all five models, AI fiction clusters in a shared region of narrative space defined by explained themes, low ambiguity, and linear time, while human fiction scatters more widely.

Why it matters: the pattern connects detection to a substantive account of what AI gets wrong about narrative. Over-explanation and tidiness are not random tics; they are what you get when generation optimizes for coherence and reader satisfaction rather than for the unresolved tension and temporal layering that characterize human literary choices. The counterpoint is that these are aesthetic tendencies, not incapacities — a sufficiently prompted or fine-tuned model can produce ambiguity and nonlinearity — so the pattern describes default behavior under typical generation, not a hard structural limit.

— "StoryScope: Investigating idiosyncrasies in AI fiction", https://arxiv.org/abs/2604.03136