With early voting already underway in the 2026 midterm elections, a growing number of Americans are experimenting with a new kind of election-season companion: artificial intelligence chatbots. Rather than calling up a friend, consulting a newspaper endorsement or attending a candidate forum, some voters are now holding extended conversations with tools like ChatGPT to help sort through their ballots, according to reporting from NPR's Weekend Edition Sunday.
One of those voters is Adam Johnson, a 40-year-old graphic designer from Morgantown, West Virginia, who is studying to become an illustrator. On a September evening, Johnson sat on his porch and worked through every contest on his upcoming ballot, weighing which candidates best matched his values. Instead of talking it over with family or neighbors, he spent hours discussing the races with ChatGPT, a chatbot built on a large language model developed by OpenAI.
For Johnson, the technology is already woven into his daily routine. He relies on ChatGPT for meal planning, help with his job search and a range of household tasks. Election research, it seems, has simply become one more item on that list. After asking the chatbot to compile a list of all the races appearing on his ballot this year, Johnson pressed further, digging into individual candidates and their positions.
His experience also hints at the limits and quirks of these tools. Johnson recalled a previous conversation in which the chatbot appeared reluctant to engage with politics at all, responding with a simple "I can't he—" before trailing off, suggesting the system had been designed to deflect or limit political discussion. That hesitation, however, has not stopped voters like him from trying to use AI as a research aid.
The trend reflects how quickly generative AI has moved from a novelty into an everyday decision-making tool. In the span of a few years, chatbots have become common for drafting emails, summarizing documents, planning trips and answering factual questions. Now, with a polarized electorate facing a long and crowded ballot, some voters are treating them as a shortcut through the noise of campaign advertising, social media claims and competing news coverage.
The appeal is easy to understand. Modern ballots often include dozens of races and ballot measures, from high-profile congressional contests down to local judicial and county offices that receive little media attention. For voters who feel underinformed, a chatbot that can generate a candidate-by-candidate summary in seconds can feel like an accessible, on-demand briefing. It is available at any hour, requires no scheduling and responds in plain language.
That convenience, though, comes with significant risks. Large language models do not retrieve an authoritative, vetted set of election facts by default. They generate responses by predicting likely word patterns, which means they can produce confident-sounding statements that are outdated, incomplete or simply wrong. A chatbot asked to compare two candidates might mix up their positions, invent a policy stance or overlook a race entirely.
Another concern is the possibility of hidden bias. Because these systems are trained on vast amounts of internet text, their outputs can reflect the perspectives, emphases and gaps present in that material. Two voters asking similar questions could receive subtly different answers depending on how they phrase their prompts, or the model could lean toward whichever candidate dominates online discussion. Neither outcome is ideal in a contest where every vote counts.
There is also the matter of transparency. When a neighbor or a newspaper editorial board makes a recommendation, the reasoning and the source are usually visible. With a chatbot, the path from question to answer is opaque. Users generally cannot see which sources informed a response, whether the information was current, or whether the model was hedging to avoid controversy. Johnson's earlier experience — a chatbot declining to engage with politics — illustrates how unpredictable these guardrails can be.
For voters, the practical takeaway may be that AI works best as a starting point rather than a final authority. A chatbot can help generate a list of races, suggest questions to ask or offer a rough overview of where candidates stand. Verifying those details against official voter guides, local news outlets or candidate websites remains essential, especially in down-ballot races where accurate information is scarcer and errors are harder to catch.
The broader implications extend beyond this election cycle. If AI-assisted voting research becomes routine, it could reshape how campaigns communicate, how voters consume political information and how misinformation spreads. Election officials and technology companies alike may face pressure to ensure that chatbots used for civic purposes are accurate, neutral and transparent about their limitations.
For now, the phenomenon is still emerging, and its scale is difficult to measure. What is clear is that some voters, like Johnson, are already treating AI as a genuine part of their civic decision-making. Whether that shift ultimately helps voters navigate complex ballots or introduces new forms of confusion will depend on how carefully the technology is used — and how rigorously its answers are checked.
(0)