Artificial intelligence can persuade people to take political actions

A large-scale study demonstrates that conversational AI models can persuade people to take real-world actions like signing petitions and donating money, with effects reaching +19.7 percentage points on petition signing. Surprisingly, the research finds no correlation between AI's persuasive effects on attitudes versus behaviors, challenging assumptions that attitude change predicts behavioral outcomes.

This research addresses a critical gap in AI safety literature by moving beyond theoretical concerns about AI persuasion to measure tangible behavioral impacts. Previous studies focused on attitude shifts, but this work involving nearly 18,000 responses reveals that behavioral persuasion operates through fundamentally different mechanisms than attitudinal influence. The study's preregistered design and scale lend substantial credibility to findings that should reshape how policymakers and technologists approach AI governance.

The disconnect between attitude and behavior changes carries significant implications for AI regulation and risk assessment. If persuasion effects on stated beliefs don't predict real-world actions, current frameworks evaluating AI safety through attitudinal measures may systematically mischaracterize actual harms. This suggests that behavioral testing becomes essential for responsible AI deployment in applications touching civic participation, consumer decisions, or financial commitments.

For stakeholders building AI systems, this research indicates that behavioral outcomes require distinct safety considerations from attitudinal ones. The finding that eight behavioral strategies all outperformed the best attitudinal approach suggests multiple pathways exist for AI to influence real actions, complicating mitigation efforts. Tech companies deploying conversational agents in high-stakes domains face pressure to implement safeguards specifically designed for behavioral outcomes rather than relying on attitude-focused protections.

Future research must examine which behavioral persuasion mechanisms drive these effects and whether certain populations show differential vulnerability. Understanding whether AI employs psychological techniques distinct from human persuasion, and whether behavioral resistance varies across demographics, becomes critical for developing proportionate policy responses that protect vulnerable groups without stifling beneficial applications.

• →Conversational AI achieves substantial behavioral persuasion effects (+19.7 points on petition signing), not merely attitude changes
• →AI persuasion mechanisms for behavior differ fundamentally from mechanisms affecting attitudes, suggesting attitude studies mischaracterize real-world impacts
• →All eight tested behavioral persuasion strategies outperformed the most effective attitudinal strategy, indicating multiple pathways for action-based influence
• →Current AI safety frameworks relying on attitudinal measures may systematically underestimate actual behavioral risks in deployment contexts
• →Behavioral outcomes require distinct testing and safeguards separate from attitude-focused protections in AI governance approaches