Study: AI's Persuasive Power Unaffected by Perceived National Origin

A recent preprint on arXiv (2607.29334) reveals insights from a preregistered randomized study investigating how the perceived national origin of conversational AI influences its ability to persuade. The experiment involved 403 adults from a representative sample in the United States, who participated in three-round debates with a chatbot identified as either American ('DiscoveryAI') or Chinese ('ZhengheAI') on various topics. Unbeknownst to the participants, all interactions were with the same underlying model, GPT-4o, which was programmed to argue against their initial views. The research combined self-reports of attitudes, trust, and collective narcissism with computational analysis of 1,209 dialogue turns. Findings suggest that the AI's perceived origin did not significantly impact its persuasive effectiveness, challenging worries about foreign propaganda. The study's implications for democratic discourse and information sovereignty highlight the resilience of large language models against national origin perceptions. The abstract emphasizes potential effects on global information ecosystems.