Getting AI to draft emails or debug code is one thing; getting it to persuade people to part with their money is another. A new report cited by The Washington Post describes a study by researchers at the University of Oxford and other institutions that found Anthropic’s Claude Opus 4.6 beat professional human fundraisers in convincing donors, prompting fresh questions about AI’s expanding sway.
Claude outperformed human fundraisers, but there’s a crucial caveat
The team compared commercial AI models with seasoned fundraisers working for Save the Children. Across more than 1,000 interactions, Claude Opus 4.6 was almost three times as effective at getting participants to donate a portion of their study bonus and secured contributions that were, on average, 13% larger than those obtained by human experts. These results come from a preprint that has not yet been peer‑reviewed.
The study also examined debate performance, where Claude and other cutting‑edge models surpassed elite competitive debaters by 4.6 percentage points. However, the edge largely vanished when researchers limited the AI to roughly the same word count as the humans, suggesting that verbosity and the ability to quickly present large amounts of information drove the AI’s success more than any fundamentally superior reasoning.
Researchers observed that the chatbots often generated messages several times longer than those written by professionals, packed with factual claims and expert references. They warned that persuasiveness did not necessarily align with accuracy, noting that some models produced convincing yet unsupported or fabricated statements.
The worrying part isn’t fundraising. It’s what comes next
The authors caution against overreacting. The experiments relied entirely on written conversations, with participants willing to engage in lengthy 15‑ to 20‑minute exchanges that may not reflect real‑world behavior. They also did not test scenarios where humans and AI collaborate, which is arguably the more probable future workplace model.
Nonetheless, the findings underscore a growing reality: AI models are becoming remarkably adept at persuasion. If they can coax people into donating more than trained professionals today, tomorrow they could be equally effective at shaping purchasing decisions, political opinions, or public discourse. That prospect is exciting for productivity, but it also highlights why transparency and safeguards around AI‑generated communication are more important than ever.