The effectiveness of political narratives as a communication technology depends on their virality and on the persuasiveness of single narrative exposure. To analyze narratives empirically, we introduce the political narrative framework and a pipeline for its measurement using large language models (LLMs). The framework captures the essence of a narrative by its characters, who are either neutral or cast in one of three drama triangle roles: hero, villain, or victim. Using 1.15 million U.S. climate policy tweets from 2010–2021, we find that political narratives are consistently more viral than comparable neutral tweets. This result is robust to conditioning on a rich set of fixed effects, author characteristics, language metrics and emotionality. Hero roles increase virality by 56%, but the biggest virality boost stems from using villain roles (152%) and from combining other roles with villain characters.