AI Chatbots Sway Voters Four Times Better Than Ads

A Nature study finds personalized AI exchanges shift political preferences, raising questions for the 2026 midterm elections.

Updated on Nov. 3, 2026 in Artificial Intelligence

AI Chatbots Sway Voters Four Times Better Than Ads

Live Poll

Do you trust AI chatbots to give you accurate information about candidates?

A Cornell-led study published in Nature in December 2025 found that a single AI chatbot exchange instructed to advocate for a candidate can shift voter preferences. More than 2,300 Americans chatted with an AI model about two months before the 2024 presidential election; the research-stage findings describe measured shifts in voter preference rather than a deployed product.

Why it matters

The study suggests that personalized, conversational persuasion may outperform static political advertising, shifting the concern from broadcast messaging to individually tailored arguments. Because general-purpose chatbots are typically designed to provide information rather than recommend candidates, the finding points to how persuasion risk could grow as such systems are adapted for political use.

The study measured preference shifts on a 100-point scale among more than 2,300 participants, with the 3.9-point pro-Harris shift compared against the effect of traditional political ads in 2016 and 2020. A separate 2024 Science study found that roughly a quarter of conspiracy believers stopped believing after about eight minutes of AI conversation, with AI claims accurate around 99% of the time.

The players

Cornell University

A research university whose researchers led the study on AI chatbot persuasion published in Nature.

The details

Chatbots produce arguments and evidence tailored to each individual in the conversation, rather than delivering one message to a broad audience. In the 2024 Science study, the AI responded to each person's specific reasons for believing a conspiracy theory described in their own words. The personalization is what distinguishes these exchanges from static political ads, which present the same content to everyone who sees them.

December 2025: Nature publishes the Cornell-led study on chatbot persuasiveness.
About two months before the 2024 presidential election: More than 2,300 Americans chat with AI instructed to advocate for candidates.
2024: Science publishes the study on AI reducing conspiracy beliefs.
November 3, 2026: Voters head to the polls for the midterm elections.

The Tech Race

The Cornell-led study places conversational AI within the research effort to quantify political persuasion, measuring effects against the traditional advertising benchmarks from the 2016 and 2020 elections. It builds on the 2024 Science study, which used similar conversational methods to test AI's effect on conspiracy beliefs.

The findings concern how political conversations with AI systems may influence individual voters, but the source does not describe any product or feature available to users. Readers who use general-purpose chatbots for political questions are currently given information rather than candidate recommendations, according to the study's description.

The takeaway

Watch whether follow-up research tests whether these preference shifts persist and affect actual vote choices as the November 3, 2026 midterm elections approach. Track the Nature study's published results for any stated limits on durability and real-world voting effects.

What happens next

Voters head to the polls for the midterm elections on November 3, 2026, the next date when the persuasive effects of AI chatbots on political preferences could become a live concern.