AI chatbots can successfully convince voters to change their candidate preferences through brief conversations, with effects lasting weeks after the interaction, according to new research published in Nature. The findings raise urgent questions about artificial intelligence's role in future elections as campaigns explore using conversational AI for voter persuasion.
Cornell University researchers recruited participants in three major elections—the 2024 US presidential race, the 2025 Canadian federal election, and the 2025 Polish presidential election. Participants engaged in text conversations with chatbots trained to advocate for specific candidates, then reported any changes in voting intentions.
In the United States, one in 21 respondents who initially opposed Kamala Harris switched their support after conversing with a pro-Harris chatbot. One in 35 Donald Trump opponents shifted toward Trump after AI conversations promoting his candidacy. The effects proved more dramatic in Canada, where one in nine participants changed their voting intention after chatbot interactions—nearly three times the US conversion rate.
Gordon Pennycook, associate professor at Cornell and study co-author, attributed the stronger Canadian results to lower political saturation. "Americans are inundated with election content non-stop," he explained. The constant campaigning makes US voters harder to persuade through any single intervention.
The chatbots proved most effective when allowed to cite facts and evidence during conversations. Removing factual claims reduced persuasive effects by more than half, demonstrating that information density drives AI's persuasive advantage over traditional advertising.
Accuracy Concerns
While most chatbot statements rated highly for factual accuracy, significant discrepancies emerged along partisan lines. Chatbots advocating for right-leaning candidates made substantially more inaccurate claims than those supporting left-leaning candidates. This pattern held consistently across all three countries studied.
The models' training on vast amounts of human-written text means they reproduce real-world phenomena, including documented patterns of less accurate political communication from right-wing sources.
Democratic Implications
The study's authors concluded that AI chatbot conversations "can meaningfully impact voter attitudes" while noting uncertainty about effectiveness if deployed by actual political campaigns. "It seems highly likely that AI-based approaches to persuasion will play an important role in future elections—with potentially profound consequences for democracy," they wrote.
Fenwick McKelvey, a Concordia University communications professor, expressed concern about combining generative AI with existing voter databases that political parties have built. These databases remain exempt from Canada's privacy laws. "The lack of oversight about databases and the data they have can be now used in ways that nobody consented to," McKelvey said.
Regulatory Landscape
Canada maintains strict guidelines on advertising and persuasion tools during election writ periods, but Elections Canada confirmed few if any rules currently address AI use during campaigns. Chief Electoral Officer Stéphane Perrault has recommended changes to elections law addressing emerging AI threats, including requiring transparency markers for AI-generated communications.
Despite demonstrated persuasive power, practical barriers exist for campaign deployment. Getting voters to engage in extended political conversations with chatbots may prove challenging. The study participants volunteered for research, creating conditions unlikely to replicate in actual campaign settings where voters actively avoid political messaging.
The research team tested 20 different AI models, including popular versions of ChatGPT, Grok, DeepSeek, and Meta's Llama. The consistency of results across models and countries suggests the persuasive effects stem from fundamental chatbot capabilities rather than specific implementations.
