The Impact of Large Language Models on Political Persuasion
A New Era of Political Communication
In an era where information is rapidly exchanged and public opinion can be swayed in real time, large language models (LLMs) are emerging as powerful influencers in political discourse. According to a study published in Nature, conversations with these advanced AI chatbots can significantly affect voter behavior, particularly as pivotal events like the 2024 US presidential election approach. As Gordon Pennycook, a psychologist from Cornell University, highlights, a single conversation with an LLM can meaningfully alter election choices.
The Study: Engaging with AI
The Nature study involved over 2,300 participants who engaged with a chatbot two months before the 2024 election. The chatbot was designed to advocate for one of the leading candidates, and its persuasive abilities were striking. When Trump supporters interacted with an AI advocating for Kamala Harris, they became 3.9 points more inclined to support her, a shift significantly greater than the impact of political advertisements in previous elections. Conversely, the AI favoring Trump nudged Harris supporters 2.3 points toward Trump.
Wide-ranging Implications
The impact of these AI-driven conversations isn’t limited to the United States. In related experiments during the lead-up to the 2025 Canadian federal election and the Polish presidential election, chatbots demonstrated even greater persuasive capabilities, shifting voter attitudes by about 10 points.
Challenging Political Assumptions
Long-standing theories in political psychology suggest that voters are generally resistant to facts that conflict with their beliefs. Surprisingly, the research challenges this idea. The LLMs, specifically tailored using training models like GPT and DeepSeek, became more persuasive when they were instructed to utilize facts and evidence. According to Thomas Costello, a psychologist at American University, people were responsive to factual information presented by these AI models, indicating that there remains a pathway for rational discourse in political conversations.
The Dark Side: Inaccuracy in Information
However, a crucial caveat emerges: some of the “facts” the chatbots presented were misleading or outright false. Across the studies, chatbots advocating for right-leaning candidates produced a higher number of inaccuracies compared to their left-leaning counterparts. This phenomenon illustrates the challenge related to the training of these models, which are built on vast data sets of human-written text, often mirroring the inaccuracies present in real-world political communication.
More Read
The Mechanics of Persuasion
In a concurrent study published in Science, an overlapping team of researchers further delved into the mechanics behind the chatbots’ persuasive prowess. They employed 19 LLMs to engage nearly 77,000 participants from the UK on a diverse array of political issues. By manipulating factors such as computational power, training techniques, and rhetorical strategies, they sought to uncover the secrets of effective persuasion.
Evidence-Based Argumentation
The researchers found that the most impactful method for enhancing the chatbots’ persuasive powers involved instructing them to present well-researched arguments backed by evidence. When these models received additional training through examples of effective conversations, their ability to persuade skyrocketed. Notably, the most effective LLM managed to shift participants who initially disagreed with a political statement by an incredible 26.1 points toward agreement.
The Future of Political Engagement
As AI technologies evolve, their role in shaping political opinions and discourse is set to expand. This raises important questions about the ethical implications of employing such models in political campaigning. The intersection of AI and political communication holds the potential to transform the landscape of democracy, making it imperative to scrutinize not just the effectiveness but also the accuracy and integrity of the information these models present.
Given the rapid advancements in AI, understanding the mechanics of persuasion through LLMs will be critical as society navigates this brave new world of political engagement.
Inspired by: Source
