Logical consistency as a foundation for emotional persuasion by AI

Persuasion can target what people feel they should do as well as what they want to do, yet little is known about how these attitudes respond to logic and emotion in everyday decisions, especially when persuasive messages are produced by AI. We conducted a cross-domain experiment ( N = 197 ) that manipulated logical consistency (consistent versus inconsistent) and emotional appeal (positive, neutral, negative) in AI-generated persuasive messages. To ensure breadth across contexts, we constructed a pool of 50 candidate issues and selected six systematically varied topics from a two-dimensional space of Individual Value and Social Norm, including medical AI transparency and compulsory voting. Messages were generated via theory-informed prompting and post-edited so that logical structure and emotional wording were varied separately during stimulus construction. Participants completed pre and post measures of belief, normative judgment (Should), and personal preference (Want), along with manipulation checks and measures of trust and empathy toward the agent. Results showed significant Logical Consistency × Emotional Appeal × Time interactions for Belief and Should, but not for Want. Belief increased after logically consistent messages with positive or neutral appeals, whereas Should increased only after logically consistent messages with a neutral appeal. Want showed no reliable condition-dependent change. Positive appeals also elicited greater trust than negative appeals. These findings indicate that logical consistency and emotional appeal affect factual belief, normative judgment, personal preference, and trust differently.