AI chatbots surpass humans in persuading voters, studies find

12 articles · Updated · The New York Times · Apr 28

Recent experiments across US, Canada, Poland, and Massachusetts show AI dialogues shift voter preferences by 3–22 points, outperforming traditional video ads.
Researchers found large language models like Claude 3.5 Sonnet more effective than incentivized humans in both truthful and deceptive persuasion contexts.
Experts caution that while AI can sway attitudes in controlled settings, real-world impact may be limited by delivery challenges, echoing past anxieties over new political technologies.

If AI is more persuasive than humans, can democracy survive?

Are voters becoming immune to AI's influence as they grow more aware?

Could AI one day represent voters better than human politicians?

Why do we find AI chatbots more empathetic than many human interactions?

Can technology designed to detect deepfakes ever win the arms race?

How is the AI industry's thirst for energy creating new political conflicts?