Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
As voters across the country head to the polls for Tuesday’s midterm elections, some may turn to a newer source for information: artificial intelligence. A voter unsure about a candidate’s election promise or a ballot measure can now simply type a question and get an immediate answer from an AI chatbot, and research led by Cornell faculty suggests those conversations can matter.
A single exchange with an AI chatbot instructed to advocate for a candidate can meaningfully shift voters’ preferences, according to a study published last December in Nature.
“Sometimes that’s good,” said Prof. Gordon Pennycook, psychology, a co-senior author of the study. “We’ve had experiments where we try to get AI to debunk falsehoods, but we also have studies showing that AI can, we might call it, bunk falsehoods — get people to believe things that are false. Or in the case of political persuasion … influence them to change who they’re going to vote for.”
The Research on Chatbot Persuasiveness
About two months before the 2024 presidential election, Pennycook’s team had more than 2300 Americans talk with an AI model instructed to advocate for either Kamala Harris or Donald Trump.
On a 100-point scale, the pro-Harrison model moved likely Trump voters 3.9 points toward Harris, roughly four times the effect of traditional ads tested during the 2016 and 2020 elections. The pro-Trump model moved likely Harris voters 1.51 points toward Trump.
According to Pennycook, AI chatbots’ advantage is personalization.
“AI models are more convincing because they are much better at producing arguments and evidence that are specific to the individual in the conversation,” he said.
A political ad, by contrast, delivers one static message to everyone. “You create an ad, and it might appeal to some people, might not appeal to other people,” Pennycook said. “But in the context of these AI conversations, every single case is specifically tuned to that individual’s, you know, concerns, thoughts, whatever.”
Whether campaigns or other groups are deploying chatbots designed to sway voters in this election cannot be known. Pennycook also pointed to a key obstacle: people scroll past ads whether they want to or not, but a chatbot conversation requires someone to choose to engage.
“It’s a higher bar to reach,” he said.
A Tool for Good or Bad
The same persuasive power can also push in the opposite direction. In a 2024 Science study, Pennycook, Prof. David Rand, information science and collaborators asked people to describe a conspiracy theory they believed in, in their own words. Then AI responded to each person’s specific reasons for believing the theory.
After about eight minutes of conversation, roughly a quarter of believers no longer believed the conspiracy, Pennycook said. He added that the AI’s claims in that study were accurate around 99% of time.
But when the researchers instead prompted the AI to argue in favor of conspiracies, people’s beliefs in those theories increased by similar amounts.
“It’s a powerful tool, and like any tool, its benefit depends on how you use it,” Pennycook said. “You can use it for good or bad.”
Across the conspiracy studies, he said, the effect did not depend on political party. “Republicans update just as much as Democrats across these studies in general.”
Pennycook also stressed that asking for political information from a general-purpose chatbot is different from talking to one that is intentionally built to persuade.
“If you just ask ChatGPT for opinions about politics, it’s not going to say you should vote for this candidate. It’ll just give you information … so that’s generally beneficial,” Pennycook said. “It’s mostly when people are using them for other reasons that they become dangerous.”
What Voters Can Do
For voters, Pennycook said, the most important step is knowing who is behind a chatbot.
“You have to be really aware of what AI you’re talking to, who has the keys, and whether there’s a thumb being pushed on the scale,” he said.
Even when no one is deliberately steering it, he cautioned, an AI model is “not a neutral arbiter of facts,” because biases are already built into how these systems work. He described the skill voters need as “another form of media literacy training.”
Ultimately, Pennycook said, chatbots summarize information that already exists rather than discovering anything new about the world.
“There’s no AI journalist that’s observing the world and putting facts down,” he said. “It’s just taking the stuff that we have and then it’s giving it back to us.”
Ashley Kim is a member of the Class of 2028 studying physics and statistical science in the College of Arts and Sciences. She is a staff writer for the science and technology department and can be reached at askim@cornellsun.com.
The Sun is editorially and financially independent from Cornell University, and has been since our founding in 1880. We rely on our readers, and your support ensures that a free and fearless student press continues to report the stories that matter most — even the ones Cornell doesn’t want us to write. But independent student journalism has never been more important, or more difficult. Give at the links below to keep The Sun shining.
