AI can outperform humans in persuasive debates, new study finds

Artificial intelligence can be as persuasive as humans—and in some cases more so—during debates, according to a new study with potential implications for political manipulation and public trust.

Researchers from the Swiss Federal Institute of Technology in Lausanne conducted online experiments involving 900 participants. Each participant was randomly assigned a topic to debate, such as school uniforms or abortion, against either a human or OpenAI’s ChatGPT-4.

The study, published in Nature Human Behaviour, revealed that AI matched humans in persuasiveness overall. However, when given personal details like age, gender, or political views, the AI outperformed human opponents 64% of the time.

Lead researcher Francesco Salvi warned that persuasive AI could be misused to microtarget voters with seemingly authentic narratives. While the AI used a more structured, analytical style, researchers found that its ability to tailor arguments to individuals was key to its impact—particularly on less polarising topics.

Experts say the findings highlight both risks and benefits. While AI could help reduce polarisation or promote healthier behaviours, concerns remain about its use in spreading misinformation, manipulating voters, and radicalisation. Scholars urge proactive regulation to prevent misuse of persuasive AI technologies.