On a September evening, Adam Johnson, 40, settled on his porch chair in Morgantown, West Virginia, to go over his ballot for the upcoming midterm election.
Chatting through all the races and the different candidates, he worked to figure out who best represented his values. But rather than discussing it with a friend or family member, he held an hours-long discussion with ChatGPT — a chatbot powered by a large language model (LLM) owned by OpenAI, an AI research company.
A graphic designer studying to be an illustrator, Johnson said ChatGPT is a significant part of his life. He uses it for meal planning, assistance in job hunting and various other tasks around the house[1].
After prompting the AI-powered chatbot to make a list of all the races on his ballot this year, Johnson dug deeper.
Johnson is one of a growing number of voters using chatbots to research their ballots for the first midterm election since AI has taken the country by storm. About half of American adults report using AI-powered chatbots and about 42% of users said they use it to search for information, according to the Pew Research Center.
Several voters NPR spoke with said they rely on chatbots for down-ballot races with thinner news coverage, to make tables comparing candidates for governors’ races and to verify or debunk viral claims. Some have been using it to create tools to help others look up their own ballots. Experts say the properties that make chatbots useful also make them subtly persuasive and eager to please.
“It might select some things that would reinforce and persuade you even more towards the way that you were leaning already,” said Rafael Batista, a postdoctoral fellow at John Hopkins University, who studies the societal and behavioral impacts of AI. “So you leave more confident, without necessarily learning more about the world.”
Having spent the vast majority of my life in states where the Secretary of State’s main job is administering elections and therefore sending out comprehensive voter’s guides four to six weeks ahead of (at least general) statewide elections, I was alarmed to discover Texas does no such thing.
That’s left to the League of Women Voters, and, well, candidates are less than enthusiastic about responding to a nonprofit that will post on a website than making sure their statements are included in something mailed out to every single registered voter.
In states with main-in voting, such as Oregon, the tempo was whatever you wanted it to be. Generally, the voter’s guide would arrive a week or two before my ballot, which itself arrived three to four weeks ahead of the deadline to return my ballot.
Some years, I’d get through most of it in a single sitting, but when you’re dealing with eight propositions, two of which are sock puppets that look similar to two placed on the ballot in good faith, things can take a bit more time. And this is where the pro/con arguments people can pay to place past the text of the measure and the SoS interpretation of impacts including financial costs and benefits.
That’s real information presented to inform. ChatGPT is neither.
maybe this is a valid use case, if buying groceries is a bit too confusing ↩︎
Using ChatGPT (or any LLM) for this purpose is obviously incredibly flawed and very likely to lead towards bias (both intentional by the service provider and unintentional from the training), and in a world of mass misinformation, a chatbot is going to be no better at filtering the facts from the bullshit than a person would be, and we know how good people are at that.
But if there’s one silver lining: anybody who did no research before who now does this at least is exposed to more perspectives, hopefully. One would imagine this is a possible gateway to further self-directed research.
Just to have a bit of fun, I sent that link to ChatGPT to see what their take would be. After the requisite throat-clearing absolving itself while agreeing it’s an inappropriate use case, I got down to brass tacks:
Following election news is admittedly a pain in the ass, especially in an environment where empirical facts now often take a back seat to the letter after a candidate’s name, and shouting nonsense is often the core of a campaign. Take the Texas Senate race as an example: If you only watch Fox News, James Talarico is a radical godless communist whose main goal is to turn all kids trans before they even make it to high school, and the campaign ads in between will confirm the narrative, when the reality is Talarico is a Presbyterian seminary student who may know a bit more about Christianity than most voters who claim to be staunch Christians. Evangelicals hate when you point out the basis of the faith is loving each other.
And this, I shit you not, was the beginning of the response:
Following an election properly is a research problem with a continuously changing corpus. You have to keep updating the set of facts you’re working from, distinguish primary evidence from somebody’s characterization of it, and notice when something that seemed peripheral becomes important.
And the Talarico example is almost comically good for demonstrating why “relevance” isn’t a neutral filtering operation.
There is a real, documented religious story here. Talarico attends Austin Presbyterian Theological Seminary and is training to become a Presbyterian minister; the Texas Tribune has reported extensively on how his theology informs his politics. His central theological argument really does involve the Christian obligation to love one’s neighbour and the idea that political action is, in part, about how we treat other people.
At the same time, he has said things such as “God is nonbinary” in a legislative speech, and Republican opponents have seized on that as evidence of his supposedly radical theology. The underlying statement exists; what it means theologically is a separate question.
Well, that’s an unexpected response, not least of which because I’ve been following the race pretty closely, and I’ve not seen this quote before, which I point out.
Talarico is a Presbyterian seminarian who has explicitly made Christianity central to his political identity; he completed a master’s degree in theological studies and has described his politics through Jesus’ commandments to love God and one’s neighbor.
And “God is nonbinary” is absolutely a real Talarico quotation. But it came from a 2021 Texas House speech about a bill concerning transgender athletes, where he was making a theological argument about biblical descriptions of God. He has subsequently explained that he was deliberately being provocative and meant that God cannot be confined to human categories.
So, I ask why the hell a quote from an unrelated situation from five years ago would come in response to a question about a candidate in 2026, and it admits there’s a fair amount of sources in current conservative circles presenting it as both current and relevant, so that’s what it went with.
Great when you already know enough to question; propaganda otherwise.
I would be more worried about the intentional bias. The temptation to use that leverage is far too great to be ignored. Only a fool wouldn’t take that opportunity.
I guess better than asking Grok who to vote for… but geez.
Is it really that hard to decide?
Quick lets ask our favorite Chinese open weight model about comparing Xi Jinping to Winnie-the-Pooh.
