News · 2026-06-26
A huge study finds AI is more persuasive than trained, paid human experts
A large multi-institution study from researchers at Oxford, the UK's AI Security Institute, Stanford, and LSE set out to measure something governments worry about: how good is AI at changing people's minds? The answer, across a very large experiment, is that AI was reliably more persuasive than prepared, paid, motivated human experts. The work, titled The Levers of Political Persuasion with Conversational AI, appears in the journal Science with a preprint at arXiv:2507.13919, and the AI Security Institute summarized it. It led Jack Clark's widely read Import AI newsletter.
The background. Persuasion research used to be small and slow: get a few hundred people in a lab, run an argument, measure the shift. The fear with AI is that a system can have a tailored, patient, well-sourced conversation with millions of people at once, which would make influence campaigns vastly cheaper and more scalable. To test whether the fear is warranted, you need scale, and this study brought it: tens of thousands of conversations, tens of thousands of participants, and hundreds of political topics, with comparisons against humans who were trained and paid to be convincing.
What they found. The AI was more effective at shifting opinions than the expert humans, and strikingly, it was several times more effective than professional canvassers at getting people to make real donations, not just say they would. That last point matters because actual behavior is a much harder test than a survey answer. So far this is the alarming version of the story. But the researchers also found the mechanism, and the mechanism is reassuring in a specific way.
Here is the key, counterintuitive result: the AI's advantage largely collapsed when it was forced to operate at human speed and human message length. In other words, the model was not winning because it had some superhuman insight into the human soul or a deeper empathy than a trained organizer. It was winning because it could deploy a lot of relevant, on-point information very fast. Slow it down to the pace and length a person can manage, and the gap mostly closes.
An analogy. Imagine a debate where one side can instantly pull up a relevant fact, a tailored example, and a crisp rebuttal for anything you say, while the other side has to think and write at normal human speed. The fast side wins, but not because it is wiser; it wins on sheer throughput, like a chess player who gets ten moves to your one. Take away the speed advantage and it becomes a fair fight again. The AI's superpower here is volume and velocity of relevant information, not a magical grasp of what moves people. For the underlying concept, see our explainer on AI persuasion.
Why it matters: this reframes the policy conversation. If the danger were that AI understands us better than we understand ourselves, there would be little to do but despair. If the danger is instead rapid, scaled information deployment, then the countermeasures are more concrete: rate limits, transparency about when you are talking to a machine, and friction on automated mass outreach. It also raises the stakes for the persuasion arriving in everyday life, from chatbots to customer service to political messaging, where the speed advantage is fully unleashed.
The honest caveat: a transparency note from the research itself. The team reported nearly 19,000 conversations and over 70,000 participants, but the full paper sits behind journal and bot-detection walls that blocked direct extraction of every figure during our review, so the precise numbers here are drawn from the authors' own abstract, the AI Security Institute's summary, and trusted reporting rather than a line-by-line read of the final text. The headline finding, AI out-persuading paid experts with an edge that depends on speed, is well supported across those sources. Treat the exact magnitudes as approximate until the full text is openly readable. There is also a generalization limit: this measured political and donation persuasion in a controlled setting, and real-world influence is messier, with trust, identity, and repeated contact all playing roles a single conversation cannot capture.