The Persuasiveness of Messages Generated by Large Language Models: A Meta-Analytic Review

As large language models (LLMs) like ChatGPT become increasingly integrated into public communication, understanding their persuasive power is essential. This meta-analysis (k = 48) synthesizes up-to-date randomized controlled experiments investigating persuasiveness of LLM-generated messages to examine: (1) the relative persuasiveness of LLM- versus human-generated messages; (2) the relative effects of different types of LLM-generated messages varying in features or source labeling; and (3) the effectiveness of LLM-generated messages compared to no or irrelevant messages.