Voting is already in progress across the country for this year's midterm elections, and a growing number of Americans are taking an unusual new approach to preparing for their trip to the polls: they are asking artificial intelligence chatbots for help. Rather than relying solely on traditional sources such as voter guides, news coverage or conversations with friends and neighbors, some voters are now using tools like ChatGPT to research races, fact-check claims and even weigh which candidates best match their personal values.
One striking example comes from Morgantown, West Virginia, where Adam Johnson, a 40-year-old graphic designer who is studying to become an illustrator, spent a September evening on his porch working through his ballot. Instead of talking things over with a friend or relative, he engaged in an hours-long conversation with ChatGPT, the chatbot built by OpenAI on top of a large language model. Johnson said the technology has become a regular part of his daily routine, helping him with tasks ranging from meal planning to his job search and various chores around the house.
For the election, Johnson began by asking the chatbot to compile a list of every race that would appear on his ballot. From there, he went deeper, walking through the candidates and the offices they were seeking in an effort to determine who aligned most closely with his values. The exchange was not entirely smooth. Johnson recalled that on a previous occasion when he raised political topics, the chatbot had seemed reluctant to engage, offering only a simple refusal rather than a substantive discussion. That hesitation reflects the careful boundaries AI companies have built around elections, but it also illustrates how voters are learning to work around those limits.
The trend is not limited to a single household or a single chatbot. Across the country, voters are increasingly turning to AI-powered assistants to sort through the flood of information that accompanies an election season. Modern ballots can include dozens of contests, from high-profile congressional and gubernatorial races down to local judgeships, school boards and ballot measures that receive little media attention. For many voters, the sheer volume of decisions can be overwhelming, and chatbots offer a fast, conversational way to organize that information.
The appeal is easy to understand. A voter can ask a chatbot to summarize a candidate's positions, compare two contenders side by side, or explain what a particular ballot initiative would actually do. The interaction feels like a discussion, and it can happen at any hour, at no cost, and without the awkwardness of pressing a neighbor or relative for political opinions. For people who feel disconnected from traditional news sources or who find official voter guides dense and hard to navigate, the technology can feel like a helpful shortcut.
Yet the practice also raises serious questions. Large language models are known to occasionally generate inaccurate or outdated information, a phenomenon often called hallucination. When that happens in the context of an election, the consequences can be meaningful. A chatbot that misstates a candidate's record, invents a policy position or mischaracterizes a ballot measure could lead a voter to make a decision based on false premises. Because the responses are delivered in fluent, confident prose, users may not realize that something is wrong.
There is also the matter of how these systems are trained and what sources they draw upon. Chatbots can reflect the biases present in their training data, and they may give different answers depending on how a question is phrased. A voter who asks a leading question may receive a leading answer, reinforcing an existing preference rather than providing balanced information. That dynamic risks turning a tool for research into a tool for confirmation, narrowing rather than broadening a voter's understanding of the choices before them.
AI companies have acknowledged these risks and have taken steps to address them. Many chatbots are programmed to direct users toward authoritative sources such as state election offices when asked about voting logistics, and some decline to endorse specific candidates outright. These guardrails are intended to prevent the technology from being used to spread misinformation or to influence elections improperly. But the boundaries are not always clear, and users like Johnson have encountered moments when the chatbot seemed unwilling to engage at all, leaving them to rephrase their questions or seek answers elsewhere.
Even with those safeguards, the growing reliance on chatbots marks a notable shift in how citizens prepare to vote. Elections have long depended on a ecosystem of information: newspapers, television coverage, debates, mailed voter guides, endorsements from community organizations and informal conversations among friends and family. AI assistants are now inserting themselves into that ecosystem, sometimes supplementing those sources and sometimes replacing them. The long-term effect on civic knowledge and democratic participation is not yet known.
Some observers see potential benefits. If chatbots make it easier for busy or disengaged citizens to understand what is on their ballot, they could theoretically increase participation and reduce the number of voters who skip down-ballot races out of confusion. A voter who might once have left a judicial retention question blank could, with a few prompts, gain enough context to make a choice. In that sense, the technology could serve as a kind of on-demand civics tutor, available whenever a voter has a question.
Others worry that convenience comes at a cost. Relying on a single automated source for political information concentrates enormous influence in the hands of a few technology companies. It also may weaken the habits of critical thinking and community discussion that have traditionally surrounded elections. If voters treat a chatbot's summary as definitive, they may not seek out opposing views or verify claims with independent sources, leaving them vulnerable to subtle errors or omissions that are difficult to detect.
The experience of voters like Johnson suggests the reality is more complicated than either optimistic or pessimistic framings allow. He did not simply accept whatever the chatbot told him; he engaged in a lengthy back-and-forth, pushing for details on individual races and trying to connect the information to his own priorities. In that sense, he used the tool as a starting point for reflection rather than as a final authority. Still, his account highlights how deeply these systems have worked their way into ordinary life, from meal planning to the voting booth.
As voting continues in the midterms, the role of AI in elections is likely to grow, and with it the debate over how these tools should be built, regulated and used. Election officials, technology companies and voters themselves will all play a part in shaping those norms. For now, the phenomenon offers a snapshot of a changing relationship between citizens and information, one in which the answers to political questions may increasingly arrive not from a newspaper or a neighbor, but from a machine trained on the sum of human text.
(0)