LLM persuasion study finds national-origin labels do not change attitude shift

A preregistered randomized experiment with 403 adults from a nationally representative United States sample found that labeling a chatbot as American or Chinese did not change how much people shifted their views after debating it, according to an arXiv preprint in Human-Computer Interaction. Participants held a three-round debate with a system introduced as either DiscoveryAI or ZhengheAI on a political or non-political topic. In every condition they actually spoke with the same model, GPT-4o, instructed to argue against their initial position. Researchers combined pre- and post-conversation self-reports of attitudes, trust, and collective narcissism with computational analysis of 1,209 participant turns. Conversations produced substantial attitude changes in every condition. The nationality label affected neither self-reported attitude change nor expressed stance, concessions, counterarguing, or affect, and equivalence tests and Bayes factors largely supported those null effects. The label's only reliable footprint was lower pre-conversation human-like trust in the Chinese-labeled model, while functionality trust was unaffected. Political topics slowed stance movement toward the AI's position. Collective narcissism predicted less attitude change regardless of origin. The authors conclude that users may withhold social trust from a rival's AI yet still assimilate its arguments, so origin labeling and transparency requirements alone may offer weak protection against foreign influence operations conducted through conversational AI.