From people to prompts: Understanding susceptibility to persuasion in conversations with artificial intelligence.

Abstract

Influencing others through conversation is fundamental to social interactions. With the rise of generative artificial intelligence (GenAI), people are increasingly relying on conversational AI for guidance on consumption and lifestyle choices, from picking a cuisine to choosing a travel destination. However, can conversational AI exert enough persuasive influence to serve as a viable substitute for human opinions? Across five experiments involving real conversations, participants discussed an opinion-based topic (e.g., whether it is worth subscribing to more than one streaming service) with a conversation partner framed as either an AI or a human. Unbeknownst to them, all participants actually conversed with a large language model prompted to try to change their minds on the topic. We found that participants showed greater attitude change toward their partner's counterattitudinal position (i.e., became more persuaded) when their partner was framed as an AI (vs. a human). We demonstrate that this occurs because people tend to view AI as inherently less persuadable than humans. As a result, they become less argumentative with an AI-framed (vs. human-framed) partner, and this reduced argumentativeness, in turn, predicts greater attitude change. Consistent with this mechanism, the effect is mediated by linguistic markers of reduced argumentativeness and attenuated when people have a goal to persuade their partner, when an AI-framed partner is described as persuadable, and when the interaction format is constrained (i.e., when argumentation is precluded and thus the partner is structurally unpersuadable). These findings advance understanding of how persuasion unfolds in conversations with AI. (PsycInfo Database Record (c) 2026 APA, all rights reserved).