The Story Shapes the Agent: Narrative Priors in LLM Behavior
Yixuan Wang, James Lester, Shashank Srivastava
Why It Matters
What makes this one worth your time
Understanding the impact of narrative on LLM behavior can enhance the design of AI systems for more effective task performance across varied contexts.
Narrative framing shapes LLM behavior more than assigned personas.
Summary
The paper investigates how narrative framing influences the behavior of LLM agents more significantly than persona prompting, using structured games to demonstrate the concept of narrative priors.
Key contributions
- Identification of narrative priors that influence LLM behavior across different tasks.
- Development of a persona-selection method that enhances cross-narrative transfer.
- Experimental validation through causal interventions demonstrating the role of anchor words in persona effectiveness.
Notable insights
- Narrative priors can explain a significant amount of behavioral variance in LLMs, suggesting a deeper layer of influence beyond persona prompting.
- Causal interventions reveal that specific language in persona descriptions can anchor behavior across different narratives.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2607.18566v1 Announce Type: cross Abstract: Persona prompting is widely used to steer LLM agent behavior, yet the narrative framing of a task can matter more than the assigned persona. We isolate this effect through structural isomorphism, constructing three text-based investigation games that share the same action space, stage progression, and resource constraints while varying only task narrative: disease investigation, IT troubleshooting, and murder mystery. Across 1,890 sessions spanning 3 models and 10 personas, we identify narrative priors: systematic action tendencies activated by a task's story framing, independent of its decision structure. Narrative priors explain 5-31x more behavioral variance than persona, are consistent across model architectures, and in two of three domains are negatively associated with task success. Persona effects that do transfer across narratives arise from behavioral anchors, persona descriptions whose language maps directly onto shared actions. Causal interventions confirm this: removing anchor words from a high-transfer persona reduces cross-narrative consistency by 95%. Our framework also generalizes to a held-out fourth narrative and yields a persona-selection method that improves cross-narrative transfer. These results suggest that LLM behavior that survives narrative changes should be grounded in concrete actions rather than abstract descriptions.