AI Systems Beat Expert Humans at Persuasion, Study Finds

A new set of four preregistered experiments — spanning 18,978 conversations from 6,923 participants — found that conversational AI systems were reliably more persuasive than laypeople, winners of an online persuasion tournament, professional canvassers, and world championship debaters. The human experts chose their own topics, researched in advance, trained for hours, and were incentivized with £1,000 cash bonuses, yet AI still came out ahead.

In a follow-up study, researchers gave experts a coaching tool letting them practice against the AI that had beaten them, review past performance, and see what the AI would have said at key moments. The AI's edge persisted regardless. Only when the AI was constrained to human-level response speed and message length did coached experts manage to tie it — pointing to raw information throughput, not superior rhetoric, as the main driver of AI's advantage.

A final real-world test showed the stakes: AI was nearly three times more effective than professional canvassers from a UK fundraising firm at raising actual donations for Save the Children. The findings suggest frontier AI's persuasive capability has real behavioral consequences, raising pointed questions for political communication, campaign strategy, and platform governance.