FSU political science expert evaluates rise in voters relying on LLMs

As artificial intelligence develops, it has become an omnipresent force in our daily lives. According to Pew Research, 42% of adults in the U.S. use Large Language Models (LLMs) like ChatGPT or Google Gemini to search for information.

This growing reliance on AI for information is now shaping the 2026 midterm elections, with voters using it to research candidates and understand ballot initiatives.

Florida State University’s Lonna Atkeson is the director of the LeRoy Collins Institute and centers her work around election science, political psychology and voting rights. She explained that this shift feels logical and may actually have a positive impact on voter turnout.

“One of the hardest things for voters is figuring out what candidates in local races actually believe. People often end up relying on party labels, endorsements or other shortcuts,” said Atkeson. “Anything that helps voters better understand candidate positions could be beneficial.”

Well-informed voting often requires hours of heavy research on candidates, especially for local elections — a concept that tends to be far too time-consuming for the average citizen. Atkeson noted that using LLMs as a voting tool provides significant information in exchange for minimal effort, making it an attractive resource.

“If AI can collect information from various sources and summarize it, that significantly lowers the cost of becoming informed,” said Atkeson. “That’s why people find it useful.”

Due to the efficiency of these models, it can be easy for users to forget that they are still developing and can make mistakes. A major flaw in LLM designs is their tendency to “hallucinate” information; this occurs when the model generates plausible-sounding information that is factually incorrect or completely fabricated.

A broader concern lies in the sycophantic nature of LLM’s. According to a series of interviews conducted by NPR, several LLM users describe them as “people pleasers” and note their tendency to “push you to what you are leaning to.” While regular users may be aware of this, it presents a dangerous bias for voters. Atkeson explained that while users should remain cautious of the technology, it can still be an effective tool if used responsibly.

“I would focus on using AI as an information tool rather than asking it who you should vote for,” said Atkeson. “Instead of asking, ‘Who should I vote for?’ ask questions like, ‘How do these candidates differ on this issue?’ or ‘What are each candidate’s positions on this policy?’ Those are the kinds of questions that are most useful.”

“I would focus on using AI as an information tool rather than asking it who you should vote for.”

– Lonna Atkeson, director of the LeRoy Collins Institute

While LLMs as a voting tool can come with defects, they may not be a major cause for concern. Atkeson noted that while voters gain information from these models, it may not affect their top-of-ticket choices, as voters tend to be firm with those.

“If we’re talking about voting for president, there’s so much information now,” explained Atkeson. “But on these down-ballot offices where you actually end up going to some gated local newspaper or don’t want to watch the hour-and-a-half debate, it’s a great option for getting information.”