AI for Persuasion? How Policy Endorsement From Large Language Models Increases Citizens' Compliance Willingness in China
通过两个有争议政策的调查实验，发现大型语言模型（尤其是中文模型）的政策背书能有效提升公众支持，而专家背书效果不显著，且公众对政策科学性的评价可能是中介机制。
ABSTRACT The voluntary compliance of ordinary citizens is pivotal for the seamless implementation of public policy and the achievement of desired policy outcomes. Government agencies often rely on expert endorsement in policy communication to bolster public support. However, recent scholarship indicates a decline in the perceived credibility of experts, particularly in China. With the rapid advancement of artificial intelligence (AI), this paper investigates whether large language models (LLMs) can notably endorse public policy and influence citizens' compliance willingness. Leveraging two controversial policies in China, we pre‐registered and performed two survey experiments. The findings reveal that policy endorsement from LLMs, especially Chinese LLMs, can effectively enhance public support for controversial policies, whereas endorsement from experts shows no significant effect. Moreover, exploratory analyses reveal that the improvement in public evaluation of the scientific nature of policymaking may mediate this causal relationship. This paper sheds light on the potential implications of AI on policy communication and citizens' policy perceptions.