Election rumors, such as unverified accounts of voter fraud, have run rampant in the US in recent years. Powered by the speed of social media and artificial intelligence (AI), it can be hard for people to know what messages to trust. Now, a team led by Caltech researchers is using AI as a tool for good in the fight against election misinformation.
In a recent paper published in the journal Royal Society Open Science, the team describes an AI tool it built to develop "pre-bunking" messages, a proactive communication technique intended to prevent the spread of misinformation by warning people of false claims before they encounter them. Testing of their scalable AI-assisted framework that generates pre-bunking articles showed the system mitigated a decrease in election confidence due to rumors among sampled voters. This effect was still measurable a week later and held across party lines.
"False claims can spread widely before fact-checkers have time to respond," says Mitchell Linegar (PhD '26), lead author of the paper who is now a postdoctoral scholar at Washington University in St. Louis. "Pre-bunking gives people accurate information before exposure, making them less likely to believe these claims, potentially stopping their spread before it starts."
Linegar worked on methods to systematically counter political misinformation during his doctoral studies with R. Michael Alvarez, Caltech's Flintridge Foundation Professor of Political and Computational Social Science and co-director of the Caltech/MIT Voting Technology Project. Together with Betsy Sinclair (PhD '07), chair of political science at Washington University and a research affiliate with Caltech's Linde Center for Science, Society, and Policy (LCSSP), and Sander van der Linden, professor of social psychology at the University of Cambridge, Linegar and Alvarez sought to use large language model–generated information to help voters navigate a complicated media environment where it is tough to separate truth from fiction.
"The rise of AI has meant that fiction is now easier and cheaper to produce—while verified truthful information is still expensive to produce—so voters are facing an onslaught of election rumors, misinformation, and disinformation," Sinclair explains. "Pre-bunking with AI is one strategy we can use to fight back and to empower voters to sort out what is true."
There has been scientific consensus for many years that pre-bunking is effective, she says, but producing it at scale has been difficult because it has required a lot of human effort to produce each pre-bunking intervention. But by combining a reusable human expert–crafted prompt with verified election information, the model developed by the team could generate pre-bunking articles for new rumors quickly and without the need for further human review.
The researchers tested the model using five common myths about the 2024 US election prior to Election Day in a sample of more than 4,000 registered voters. Participants were randomly assigned to read a persuasive human-written article endorsing one of five commonly believed election myths. Some then received an AI-generated pre-bunking article addressing the myth, while others received an AI-generated article about an unrelated subject. After the intervention, the researchers measured the participants' beliefs in election myths, confidence in true election facts, and overall trust in election integrity, with follow-up measurements one week later.
"Purely AI-written pre-bunks were just as effective as those receiving human feedback, allowing us to front-load human effort and expertise and making it possible to address new rumors quickly and at low cost," Linegar says. "Our work showed that a short AI-generated pre-bunk protected voters from false election rumors."
The AI-assisted tools will be particularly helpful in responding to attempts to decrease confidence as we head into the 2026 midterm elections, Alvarez says, as deliberate misinformation efforts seeking to disenfranchise voters is a major concern. "This is really important, because AI tools can develop countermeasures quickly, which will help us get ahead of misinformation campaigns," he notes.
In addition to developing a tool that can keep up with the rapid spread of misinformation, designing the template for the pre-bunking messages was particularly challenging, Linegar says.
"A pre-bunking article has to introduce a false claim clearly enough for people to understand it without being persuasive enough that the intervention causes harm; traditionally, a human expert does this for each new rumor," he explains. "It took a lot of iteration to get the model to distill and weaken the claims without repeating them verbatim."
Alvarez says the team's success was due, in part, to their interdisciplinary approach; their collaboration with van der Linden, a world-renowned expert on pre-bunking and countering rumors and conspiracy theories; and the three generations of Caltech quantitative social science experts who helped run the study.
"I've been working with Mike [Alvarez] on studying how technology can support democratic elections since the rise of the Caltech/MIT Voting Technology Project; this is the kind of project that made me want to be a political scientist," Sinclair says. "I think this is one the ways in which Caltech has demonstrated tremendous intellectual leadership on one of the most pressing problems of our age."
To facilitate real-world use, the team has released a public demonstration of the tool that drafts pre-bunks using trusted factual material at electionbot.chat/article. They hope to work with government partners, election officials, and other trusted public communicators who can use the framework themselves to increase public trust in democracy as they see rumors beginning to emerge.
"Across the country there are myriad hard-working election officials who are answering phone call after phone call, email after email, addressing questions that have arisen because of the rise of election rumors," Sinclair says. "Tools like the one we have developed are going to make their jobs easier, while also making it easier for voters to access accurate information."
In collaboration with state officials, Linegar says the team is developing chatbots that give voters accurate information grounded in their own state's laws and procedures.
"We are also exploring shorter formats, such as social media posts and videos, and applications beyond elections," Linegar says. "The goal remains the same: help accurate information move as quickly as the misinformation it competes with."
The Royal Society Open Science paper is called "Towards scalable AI-assisted pre-bunking of election misinformation: evidence from a pre-registered US panel experiment." The work was supported by a grant to Caltech from the John Randolph Haynes and Dora Haynes Foundation.