AI systems are better at persuasion than real people, out-performing winners of an online persuasion tournament, professional canvassers and world championship debaters.

Along with being more convincing on issues from immigration controls and the monarchy to physician-assisted suicide and the conflict in Ukraine, they make better charity fundraisers than real people, getting substantially more real-money donations than well-paid professional canvassers.

In a series of four experiments examining around 19,000 conversations with nearly 7,000 people, expert humans were allowed to choose their issues, research in advance and undergo hours of live, structured practice—and were even incentivized with £1,000 cash bonuses.

Despite this, though, they still failed to convince as often as AI, with AI nearly three times as effective than professional canvassers from a UK fundraising firm when it came to raising real-money donations for Save the Children.

"On average, AI exceeded every class of human persuader we tested: random laypeople, tournament-selected laypeople, and even elite debaters," the Oxford University researchers said.

And in a follow-up study, AI's advantage persisted even after experts were given a coaching tool that let them practice against the AI that beat them, review their performance history, and see what AI would have said at key moments.

The AI's advantage, the researchers found, appeared to stem from its ability to make use of larger quantities of information at high speed. Once they'd been coached, expert humans could match an AI as long as it was forced to respond at human speeds and with human-length messages.

The findings could have significant implications, said the team, allowing political campaigns, lobbying firms or fundraising drives with access to the most capable AI to prevail over rivals with stronger cases or more talented human advocates.

Meanwhile, those looking to spread disinformation, target individuals with coercive personalized influence campaigns or perpetrate scams could find their job made easier.

"Our findings establish frontier AI as a more capable conversational persuader than the most prepared, incentivized, and expert humans we could recruit. Training humans does not appear to close that gap," the researchers said.

"As access to these systems continues to grow, the question is no longer whether AI can out-persuade humans but how, where, and on whose behalf this capability will be exercised."

The results echo a similar study carried out last year by the AI Security Institute, which found that information density drives persuasion more than personalization or microtargeting.

But, said the team, more persuasive models made more inaccurate claims—although this doesn't mean that inaccurate claims are more persuasive than accurate ones.

"One possible explanation for this correlation is that as models become more information-dense in their responses, they eventually exhaust the supply of reliably accurate facts and begin to generate weaker or less accurate material," the researchers said.

"The same techniques that increase persuasiveness also reduce factual accuracy, highlighting the importance of robust evaluation standards, transparency around model training, and careful consideration of any optimization that might affect influence or truthfulness."