AI Out-Persuades Expert Humans in Massive Study, Raising Societal Power Questions

A large-scale study across nearly 19,000 conversations found that current AI systems consistently outperform expert human persuaders in text-based debates and fundraising—even when those humans are elite debaters given extensive coaching and cash incentives. The research, a collaboration between Oxford, the UK AI Security Institute, Stanford, and the London School of Economics, tested persuasion on policy issues and real-money donations to charity.

In one experiment, AI was nearly three times more effective than professional canvassers at raising actual donations for Save the Children. The leading models were Anthropic's Opus 4.1 and 4.6, followed by systems from OpenAI, Google, and xAI. When researchers constrained AI to human message lengths and typing speeds, its advantage over coached human experts vanished—suggesting the edge comes from information density and processing speed, not inherent rhetorical superiority.

The findings highlight a societal fork. Cheap, scalable AI persuasion could empower under-resourced actors like public defenders or grassroots activists. But it could also consolidate influence among those who already control powerful AI systems, potentially supercharging advertising, political campaigns, and propaganda. The researchers warn that how we allocate and regulate these persuasive capabilities will shape the balance of power going forward.