Narrative Feature Analysis
Paste a story or creative text on the left and click Analyze.
Based on 30+ discourse-level features: temporal structure, thematic explicitness, moral ambiguity, reader engagement, and more. Uses the NarraBench taxonomy from the StoryScope paper by Russell et al. (2026).
Narrative Space: Approximates Figure 2 from the paper. LD1 separates human (left, negative) from AI (right, positive). Your text (★) is plotted based on extracted feature scores against the background of 200 synthetic points per source calibrated to the paper's reported centroids.
Rarity Violin: Approximates Figure 5. Human stories occupy less-crowded regions of narrative space (mean rarity 0.71 vs 0.49 for AI). The dashed line shows your text's estimated rarity percentile.
| Feature | Value | Human | AI | Humanness | Signal |
|---|
Rewrites the text targeting its most AI-elevated narrative features: structural choices, not just surface style. Re-analyzes after to show movement in narrative space.
Paste as system prompt to steer AI generation toward human-like patterns.
Select constraints, then build a prompt snippet to prepend.
Why surface rewrites fail: The LAMP framework (span-level surface editing) drops detection accuracy by only 1.6pp (93.9% vs 95.5%). Surface edits don't change the underlying structural choices that drive detection.
5 strongest AI signals (largest gap):
- Emotional expression via embodied metaphors (AI=81%, Human=38%)
- Thematic explicitness & moralizing (AI=3.94, Human=3.28)
- Setting as psychological mirror (AI=4.07, Human=3.58)
- Moral/philosophical weighting (AI=3.68, Human=3.26)
- Environmental emphasis (AI=3.21, Human=2.83)
5 strongest human signals:
- Moral polarity → ambivalent (Human=59%, AI=38%)
- Intertextual named references (Human=47%, AI=24%)
- Fourth-wall permeability (Human=0.67, AI=0.39)
- Depth of recontextualization after surprise (Human=3.28, AI=2.95)
- Chronological discontinuity (Human=2.40, AI=2.12)
Model fingerprints: Claude → flat escalation + epilogues. GPT → gossip/dreams + retrospective narrator. Gemini → protagonist social expansion. DeepSeek → interleaved backstory. Kimi → in medias res.