When Meta’s AI chatbot was asked in Spanish for the date of the midterm elections in the US, the answer that came back was November 4 — not November 3. The same incorrect response was returned twice, said Laura Zommer, co-founder of Factchequeado, a fact checking site combating Spanish-language misinformation.
When the Institute for Strategic Dialogue put the same election questions to six chatbots in English and Spanish, about 3 in 10 English answers were incomplete, outdated or wrong. That compares to 45% in Spanish, Zommer said, speaking at an American Community Media national briefing on October 23.
AI has become a critical player in the voting process, accelerating the rapid spread of misinformation and disinformation. From language inconsistencies to unsubstantiated claims and misleading narratives, information around elections is becoming harder to sift through.
Using AI to research candidates
Voters are increasingly relying on AI for election questions raising concerns about information accuracy.
Cameron Hickey, CEO of the National Conference on Citizenship (NCoC), said that their polling has shown that the share of eligible voters using a chatbot to research candidates has been rising as the midterms approached.
“We asked that question two weeks ago and the answer nationally of eligible voters was 14%. We asked it again this week and it went up to 25%,” he said, adding that for non-Hispanic eligible voters, reliance on AI hovered at 13%, but for Hispanic voters that number shot up to around 30%.
Laura Zommer, Co-founder and CEO of Factchequeado, discusses the use of AI in political ads and how the federal and state governments regulate it.
Stoking fear
To drive the continued interest in using social media platforms, advanced AI models — the same used to power social media algorithms — populate social media feeds with messages, “and those often are the mis- and disinformation messages that enrage us or breed fear,” said Hickey.
He pointed to one video in which ICE agents are seen intimidating voters at polling sites. The voters in the video are fake, said Hickey. But the images are convincing, in part, because ICE has already been spotted at polling places, while administration officials have so far declined to rule out the possibility of deploying agents to polling sites on or before election day.
Such a deployment would violate both federal and state law with few exceptions.
Other videos show touchscreen machines switching votes, and “this exact claim has been around for the last three election cycles,” said Hickey, adding years of investigation have shown this to be a highly unlikely occurrence.
Cognitive biases, meanwhile, influence the reactions elicited by these messages.
Hickey gave the example of a recent rumor that automated Russian bots supposedly decreased dramatically after Ukrainian strikes on Russian data centers. The rumor took hold on social media before Ukrainian President Volodymyr Zelensky validated it at the Warsaw Security Forum in mid-October, saying there was “less hate, less division and fewer provocations” online because of fewer bots.
Part of this claim was true. There were strikes on Russia’s data centers. But “in fact, the same number of bots were there the next day as the day before,” said Hickey, noting the rumor “spread far and wide quite quickly because people kind of wanted that to be true.”
Targeting Asian voters
Jinxia Niu, Chinese Digital Engagement Project, Chinese for Affirmative Action, discusses misinformation and why certain groups are more likely to believe it than others.
Jinxia Niu of Chinese for Affirmative Action, which runs the fact-checking site PiYaoBa.org, said that in the Chinese community, the nexus of disinformation has shifted from WeChat to X.
“But WeChat is still very important because disinformation usually starts with X but then gets translated and circulated in WeChat groups and YouTube,” she said.
Younger people are more adept at distinguishing misinformation, continued Niu, explaining that immigrant seniors are more vulnerable, especially if they are highly educated. “The more educated they are, they tend to stay online on social media more often with more exposure to all this disinformation.”
Along with PiYaoBa, VietFactCheck, Tayo and Savera are reliable sites to verify claims for other Asian communities.
The virtually nonexistent, noncitizen voter
Election fraud is the biggest theme of disinformation this year, with noncitizen voting claims exploiting cultural differences, said Niu.
Chinese immigrants on H1B visas, international students and green card holders have been asking on WeChat groups and Xiaohongshu — a Chinese social media and e-commerce platform — if their immigration applications could be affected if they were added to the voter rolls when applying for a driver’s license, whether due to a software error, or an incorrect automatic selection.
Hickey pointed to claims about noncitizen voting proliferating and “being amplified by what would otherwise be considered trusted messengers, i.e. our own government.”
An internal report by Homeland Security Investigations, an arm of ICE, on noncitizen voting, found that “the phenomenon is virtually nonexistent,” wrote Jonathan Blitzer in the New Yorker.
Extensive research also indicates that these votes have never affected the outcome of an election. Yet the Trump administration remains fixated on illegal voting, with a USCIS quota to investigate 40 names per day.
According to Zommer, fighting disinformation starts with listening and paying attention to better understand what the real conversations are. Hickey agrees, advocating for what is known as the SIFT method, a framework for assessing the accuracy of online information. The letters stand for Stop, Investigate the source, Find better coverage and Trace claims.
JAYA PADMANABHAN
ACoM Executive Dir. Jaya Padmanabhan is an Emmy Award-winning editor and veteran journalist with decades of experience across newsrooms and ethnic media. Her freelance work has appeared in the New York Times, PBS Next Avenue, Forbes, and other publications.
