Scores of political campaigns are now using generative artificial intelligence to talk to thousands of voters, sometimes without disclosing that the contact is coming from AI. The chatbots can effectively mimic human campaigners and engage in limitless conversations through text messages and robocalls. The technology allows campaigns to conduct more personalized outreach to more people, but it also comes with increased risks of spreading inaccurate information. It also has the potential of taking data collection on voters to a whole new level.
Currently, campaign-deployed chatbots are largely limited to mobilizing voters and to touting campaign messages that are more targeted to individual voter interests and demographics than ever before.
Many observers focus on the fact that generative AI tools are plagued by “hallucinations” — made-up scenarios or sources presented as fact. Several studies conducted around the 2024 elections in the United States and overseas found that popular generative AI tools sometimes failed to provide voters with accurate information about candidates and policies.
Generative AI chatbots have well-documented tendencies to default to flattering and appeasing their users. In the context of political campaigning, that means chatbots may be inclined to craft messages that appeal to voters’ preferences or validate voters’ assumptions, even if those messages are misleading or flat-out false.
To some degree, that’s nothing new: As early as 64 BC, the Roman orator Cicero’s brother advised him to promise everything to everyone during a political campaign, and countless politicians have adopted that credo ever since. But when it comes to interactive, generative AI-powered conversations with voters, the absence of any real-time vetting by campaign staff erases a layer of accountability that might provide some assurance that the campaign actually endorses a particular message.
The public has so far paid less attention to AI tools’ potential for political persuasion, including both how the tools could be deliberately exploited by bad actors or harnessed for good by pro-democracy forces. Scholars, however, have been examining these questions for some time, and their investigations have yielded intriguing results.
Following the generative-AI boom in 2022, social scientists began testing chatbots’ persuasive capabilities. Several of these studies explored chatbots’ pro-social capabilities, such as their ability to combat falsehoods and conspiracies. Two studies published in recent years, for example, suggested that pushback or incisive questioning by chatbots could reduce some users’ belief in conspiracy theories for a sustained period.
The results surprised scholars, because research had previously shown that conspiracy theory beliefs are relatively resistant to change, especially where such beliefs are tied to partisan identity and distrust in institutions. The AI chatbot study results indicated that prior interventions may have failed because of brevity or generic messages, whereas chatbots powered by large language models could overcome these limitations by delivering more in-depth responses tailored to conspiracy theory believers’ specific rationales and identities and by engaging in persistent conversation.
But this same promise for pro-social use could translate into danger in the voting context. Generative AI chatbots could work as a kind of scalable persuasion infrastructure for partisan or malevolent ends. The same study that found debunking could successfully rebut conspiracies, for example, also found that instructing OpenAI’s GPT-4o to argue in favor of conspiracy theories can increase the beliefs of exposed users in those theories. A tool that can be mobilized to help defeat election lies can also be used to bolster them.
Similar investigations support that conclusion. For instance, one recent study found that chatbots powered by large language models can influence preferences around candidates and political issues, including by promoting falsehoods. The study examined AI persuasion in the 2024 U.S. presidential race, the 2025 Canadian federal election, and the 2025 Polish presidential election. Participants in the study conversed with chatbots advocating for a major candidate. The study found that chatbot conversations could strengthen preferences for candidates, and the effects of such persuasion were larger than the typical measured effects of televised political ads. Notably, in addition to supporting arguments with relevant facts and evidence, the chatbots sometimes relied on false claims to try to sway voters.
Other scholars have produced work illustrating parallel risks. Strikingly, one study showed that AI tools could sometimes outperform humans in online debates. Together, the emerging research on generative AI conversations suggests that the technology has strong persuasive potential that will likely increase as the tools become better at selecting evidence, tailoring responses to users, and countering rebuttals.
How are campaigns using AI-driven conversations today, and how could that transform in the future? In the 2026 election cycle, some campaigns used AI-powered chatbots to mobilize voters and promote campaign messages. In the future, less scrupulous campaigns or political committees could use chatbots to discourage turnout from opponents’ supporters by smearing other candidates and promoting cynicism and distrust in elections and democracy. Such conversations may become even more common through large language model–powered phone calls and emails, QR codes in mailers, and ads that link to campaign or PAC-affiliated online chatbots.
