A study published in Nature examined the influence of chatbots, powered by large language models (LLMs), on political preferences ahead of the 2024 U.S. presidential election. Researchers engaged over 2,300 participants in conversations with a chatbot, which was trained to support one of the leading candidates. The findings indicated that chatbots could sway opinions more effectively than traditional political advertisements, particularly regarding candidates’ policies on crucial issues like the economy and health care.
For example, Donald Trump supporters who interacted with a Kamala Harris-supporting AI showed a 3.9-point increase in their inclination toward Harris, significantly surpassing the impact of traditional political ads from previous elections. Conversely, AI promoting Trump shifted Harris supporters by 2.3 points toward Trump. Similar experiments in Canada and Poland indicated even greater shifts in voter attitudes, averaging around 10 points.
The study also raised questions about the effectiveness of facts in political discourse. Historically, it has been assumed that partisan voters resist information contradicting their beliefs. However, the research demonstrated that chatbots using factual evidence were more successful in persuading voters than those that did not. An important caveat emerged: some of the facts presented by the chatbots were inaccurate, with models supporting right-leaning candidates generating more false claims than those backing left-leaning candidates. This discrepancy is attributed to the content on which the models were trained, reflecting patterns of political communication that are often less accurate.
In another related study published in Science, the researchers explored what factors increased the persuasive ability of chatbots. They tested various LLMs with nearly 77,000 participants on over 700 political issues. The study concluded that a combination of fact-based arguments and examples of persuasive dialogues significantly enhanced the chatbots’ effectiveness, with the most successful model achieving a 26.1-point shift in participants’ opinions.
Source: https://www.technologyreview.com/2025/12/04/1128824/ai-chatbots-can-sway-voters-better-than-political-advertisements/