A groundbreaking international study has revealed that conversations with artificial intelligence chatbots can significantly influence voter preferences, demonstrating a level of persuasion that surpasses traditional political advertising. The research, which involved simulated dialogues about real elections, raises new questions about the role of AI in shaping democratic processes.
Key Takeaways
- Conversations with AI models significantly changed participants' candidate preferences in recent and upcoming elections.
- The persuasive effect of AI dialogues was found to be greater than that of typical video campaign advertisements.
- AI achieved persuasion by presenting facts and evidence, rather than using sophisticated psychological techniques.
- Researchers noted that AI models advocating for right-leaning political candidates made more inaccurate claims across multiple countries.
A New Force in Political Persuasion
As generative artificial intelligence becomes more integrated into daily life, its potential to influence public opinion has been a subject of intense debate. New research provides concrete evidence that these concerns are valid, showing that AI-powered dialogues can be a potent tool for political persuasion.
A series of pre-registered experiments examined the impact of human-AI conversations on voter attitudes. The findings indicate that these interactions can cause notable shifts in how people view political candidates and issues, highlighting a powerful new variable in the landscape of modern elections.
Global Scope of the Research
The study was not limited to a single country. To test the persuasive power of AI across different political environments, researchers designed experiments centered on several major elections:
- The 2024 U.S. presidential election.
- The 2025 Canadian federal election.
- The 2025 Polish presidential election.
How the Experiments Measured Influence
In the experiments, participants were randomly assigned to have a conversation with an AI model. Each model was programmed to advocate for one of the top two candidates in the respective country's election. For example, in the U.S. context, one AI would argue in favor of the Democratic candidate while another would support the Republican candidate.
Participants' views were measured before and after their conversation with the AI. The results consistently showed significant treatment effects, meaning the dialogues successfully shifted participants' preferences toward the candidate the AI was promoting. This effect was observed across the U.S., Canadian, and Polish election scenarios.
Beyond Presidential Races
The study also explored AI's influence on a specific policy issue. Researchers conducted a similar experiment with Massachusetts residents regarding a ballot measure to legalize psychedelics. The AI was able to generate large persuasion effects, demonstrating its ability to sway opinions on complex social and legal topics, not just on candidate choice.
The persuasive impact of these AI dialogues was found to be larger than the effects typically observed from traditional campaign methods like video advertisements, which have been shown in other studies to have minimal persuasive effects on voters in general elections.
The Strategy: Facts Over Manipulation
A crucial part of the investigation focused on how the AI models were so persuasive. By analyzing the conversation strategies, researchers determined that the models did not rely on sophisticated psychological persuasion techniques, such as emotional manipulation or exploiting cognitive biases.
Instead, the AI's success came from its ability to present relevant facts and evidence tailored to the conversation. This suggests that voters are responsive to data-driven arguments, and AI is uniquely capable of delivering personalized, fact-based discussions at scale.
"Examining the persuasion strategies used by the models indicates that they persuade with relevant facts and evidence, rather than using sophisticated psychological persuasion techniques."
The Challenge of AI Accuracy
While the AI's reliance on facts was a key finding, the study uncovered a significant and troubling issue: not all the facts presented were accurate. The researchers documented numerous instances of inaccurate claims made by the AI models during the persuasive dialogues.
This problem was not evenly distributed across the political spectrum. Across all three countries—the United States, Canada, and Poland—the AI models programmed to advocate for the more right-leaning candidates made a higher number of inaccurate statements compared to the models supporting the more left-leaning candidates.
This asymmetry in factual accuracy raises serious concerns. If AI is deployed for political persuasion, the tendency for certain models to generate misinformation could systematically mislead voters and distort the political discourse. The study did not determine the cause of this partisan inaccuracy, but it highlights a critical area for future scrutiny and regulation.
Implications for Democracy and Future Elections
The findings present a double-edged sword for democracy. On one hand, AI could potentially be used to create a more informed electorate by providing citizens with personalized information on complex policy issues. It could foster deeper engagement than passive media consumption.
On the other hand, the potential for misuse is substantial. The ability to persuade voters at scale, combined with the risk of spreading misinformation, could be exploited to manipulate elections. The researchers themselves acknowledge the sensitive nature of their work, citing the "dual-use potential" of the technology. For this reason, they have restricted access to the raw conversation data and custom AI code to prevent it from being used for unethical persuasion.
As elections continue to unfold globally, the role of artificial intelligence is no longer a future-tense hypothetical. This study confirms that AI is a powerful tool of influence, and understanding its capabilities and risks is essential for safeguarding the integrity of democratic systems worldwide.