AI detects AI fiction by how stories are built, not how they sound

Researchers built a system that identifies AI-written stories by analyzing narrative structure — how characters make decisions, how time flows, how plots branch — rather than looking at word choice or sentence patterns. This means AI detection can work even if an AI learns to mimic human writing style, because the underlying story architecture stays recognizably different.

For years, AI detection relied on surface-level fingerprints: sentence length, vocabulary, punctuation. Those break the moment an AI learns to imitate. This paper shows AI has a deeper signature — the way it constructs narrative itself. AI stories tend to over-explain, favor single-track plots, and make character choices seem inevitable rather than genuinely uncertain. Human stories are messier and more temporally complex. The practical effect: if this holds up, detection becomes much harder to game. The unstated implication: AI writing isn't just stylistically different, it's structurally different in ways that might be hard to fix without rethinking how these models generate narrative.