StoryScope Reveals the Narrative Fingerprints of AI Fiction

StoryScope is a research pipeline that distinguishes AI-generated fiction from human writing by looking past surface style toward discourse-level narrative choices, such as character agency and chronological discontinuity. The system extracts interpretable features across 10 dimensions and was applied to a parallel corpus of 10,272 writing prompts, each written by one human author and five different LLMs, producing 61,608 stories with 304 narrative features extracted per story.

Using narrative features alone, the pipeline reaches 93.2% macro-F1 for human-vs-AI detection and 68.4% macro-F1 for six-way authorship attribution, retaining over 97% of the performance of models that also include stylistic cues. A compact set of 30 core features captures most of this signal: AI stories tend to over-explain themes and favor tidy, single-track plots, while human stories frame protagonists' choices with more moral ambiguity and greater temporal complexity.

Model-specific fingerprints stand out as well: Claude tends toward flat event escalation, GPT over-relies on dream sequences, and Gemini defaults to external character description. AI-generated stories cluster tightly in a shared region of narrative space, whereas human-authored stories show far more diversity. The findings suggest that differences in underlying narrative construction, not just writing style, can help separate human-written fiction from AI-generated text.