Zero-Shot Narrative Detection via Human-Written Descriptions
September 16, 2026
LLMs can identify hidden strategic narratives in social messaging using zero-shot prompting with human-written descriptions. The study found that human descriptions outperform few-shot or automatically generated descriptions, which can degrade accuracy due to framing shifts.
HOW THIS AFFECTS YOU
●
builderYou can leverage large models for narrative analysis by using high-quality human descriptions instead of few-shot examples.
●
researcherThis highlights the sensitivity of LLM pragmatic reasoning to the framing of descriptive prompts.