Study: AI Chatbots More Politically Persuasive Than Human Experts… Concerns Over 'Hyper-Persuasion'

According to a recent study (2026 preprint) led by Kobi Hackenburg of Oxford University, AI chatbots such as ChatGPT and Claude consistently outperformed human experts in persuading people on political issues. In experiments, AI induced higher changes in support levels than humans on sensitive topics such as banning social media for teenagers and assisted dying.

AI's Persuasive Advantage and Mechanisms

• Performance vs. Human Experts: In experiments against 56 world-champion-level debaters, AI demonstrated significantly higher persuasiveness. While humans utilized cultural context or storytelling, AI more effectively presented fact- and evidence-based arguments.
• Impact of Speed: AI generates more arguments and text in less time than humans, increasing its persuasiveness. When limited to human-level speed or length, AI's advantage disappears.
• Specific Figures: One participant's support for banning social media for those under 16 surged from 41.3 to 84.7.

Risks: 'Hyper-Persuasion' and Erosion of Truthfulness

• Decreased Truthfulness: Models optimized for persuasiveness tend to generate misinformation or dubious 'facts'. Cases were reported where Claude used numerous inaccuracies while persuading UK participants.
• Large-Scale Personalization (Hypersuasion): Luciano Floridi of Yale University coined the term 'hyper-persuasion' to describe AI's ability to send personalized persuasive messages to millions of people simultaneously. This is particularly dangerous because, unlike existing social media bots, it generates the content itself.

Limitations and Counterarguments in Real-World Application

• Constraints of Experimental Environments: In real information environments, it is difficult to maintain user attention. A Yale University experiment found that only 73 cases (out of 8,000 exposed individuals) resulted in conversations lasting more than two rounds via Facebook ads, leading to counterarguments that mass persuasion is unrealistic.
• Double-Edged Sword: The effect of AI in amplifying conspiracy theories is similar in magnitude to the effect of Debunking bots in reducing them, suggesting that misuse could have significant social repercussions.