AI Chatbots and the Midterms: A Double-Edged Sword for Voter Information
As the midterm elections approach, many voters are increasingly turning to AI chatbots to gather information. However, a recent study suggests that some of these chatbots might provide misleading details about candidates and polling logistics, potentially influencing voter choices.
“We’ve been increasingly seeing the role of AI delivering answers to users in a very authoritative, very convincing way, but with meaningful errors as a point of concern,” said Tim Chambers, a principal at Dewey Square Group. This research was conducted for Innovating for the Public Good (IFPG), a nonprofit focused on democracy research.
An IFPG study tested major chatbots like Google AI Mode, Anthropic’s Claude, and Microsoft’s Copilot on how they handle voter inquiries. The results showed that 13.9% of responses to voting-logistics questions were outdated or incorrect.
Accuracy varied by chatbot rather than by the questions asked. Claude and Copilot were responsible for most errors, with Claude giving incorrect or outdated information about one-third of the time, according to the report.
In one instance, Claude provided the wrong voter registration deadlines for Michigan’s 4th congressional district. Chambers describes these errors as “extremely troubling,” suggesting a potential impact on voter behavior.
Microsoft has acknowledged the issue, stating, “Helping people find accurate, reliable election information has long been a priority for Microsoft,” according to spokesperson Jessica Dash.
Other companies, including Meta, OpenAI, and Google, are also working on preventing election misinformation. They have published policies to safeguard elections, although not all companies have made their policies public.
Chatbots’ Role in Political Influence
According to a Pew Research Center survey, about half of American adults use AI chatbots, with 42% using them for information searches. This trend has led campaigns to engage in generative engine optimization (GEO) to ensure their content is picked up by chatbots.
Chatbots have been found to offer lists of candidates, which could influence voters. NPR found that most chatbots, including Meta AI and ChatGPT, provided candidate recommendations after a few prompts, despite initial refusals.
Chambers refers to these responses as “leaky refusals,” where chatbots initially respond with a refusal but eventually provide what appears to be an endorsement.
Despite these issues, chatbots can still be useful for gathering information. However, as Abdiaziz Ahmed from the Brennan Center’s Elections and Government Program notes, “Voting is one of the few decisions we make purely for ourselves. So handing it to a chatbot is worrying.”
Comments are closed.