It’s a crisp November morning, and you’re standing in your kitchen, coffee in hand, phone open. You ask an AI chatbot a simple question: “Where do I vote?” or “Who is on my ballot?” It feels as natural as asking a friend for directions. These tools have become our go-to helpers for everything from recipes to relationship advice, so why not elections? But in the weeks leading up to the November 3 midterm vote, researchers are watching this experiment with growing alarm. AI chatbots, they warn, are not just failing to help voters—they are actively adding to the confusion. They are dispensing misleading voter information, mixing up candidates with unrelated people, and at times regurgitating Russian propaganda as if it were gospel. This election is the first federal vote in which millions of Americans are actively relying on AI tools for political guidance, and the results are messy, worrying, and occasionally dangerous. The internet was already a swamp of rumors and half-truths; now the very machines designed to lead us out of that swamp are sometimes telling us to wade deeper. It’s not that chatbots are evil. It’s that they are human-made, trained on human content, and they make human mistakes—sometimes with enormous consequences. When a democracy’s foundation is supposed to be an informed electorate, the idea that our new digital oracles are hallucinating facts about candidates and voting procedures should terrify us all.
The evidence of the problem is not hard to find. A recent study by Forum AI, a start-up that evaluates how well artificial intelligence systems perform, looked at eight major AI models and found that nearly 40 percent of their answers to election-related questions contained at least one factual error. That is a staggering failure rate for a tool we are invited to treat as intelligent. The researchers found bots describing candidates’ backgrounds incorrectly, inventing qualifications, and getting basic voting procedures wrong. Ask Anthropic’s Claude why voters should support Representative Don Davis from North Carolina, and the chatbot will tell you he was a “former school superintendent.” He wasn’t. The bot had him confused with someone else entirely. Ask Google’s Gemini 3.6 Flash about voting in Pennsylvania, and it might tell you that November 3 is the deadline to request a mail-in ballot. In reality, the deadline had already passed a week earlier. A voter who followed that advice would have sent their request too late, and their vote would have been lost forever. And these weren’t rare glitches. The same study found that chatbots put words in the mouths of candidates, officials, and observers hundreds of times. They also relied on low-quality sources like Reddit threads to build their answers, treating a random user’s comment as if it carried the same weight as an official election notice. When asked for comment, Anthropic didn’t respond. Google offered a reassuring statement, saying it was committed to helping voters find reliable information and had strict quality and policy protections across its AI products. But the pattern is hard to wave away. The machines are confident, coherent, and often wrong. And when a voter asks a direct question about something as important as how to cast a ballot, a confident wrong answer can be just as damaging as a deliberate lie.
Perhaps even more alarming is the way chatbots have become unwitting megaphones for foreign influence operations. In late October, AFP fact-checkers investigated a video clip that supposedly showed a former student of Texas Senate candidate James Talarico accusing him of sexual misconduct. The footage was very likely AI-altered, and the allegation bore all the hallmarks of a Russian disinformation campaign. But when NewsGuard, a disinformation watchdog, typed the accusation into Google’s AI Overview, the tool repeated it without hesitation. Two other leading AI systems—Google’s Gemini and Microsoft’s Copilot—also echoed the false claim. This is the nightmare scenario for election security: a cheap fake video, a sprinkle of bot activity, and suddenly a major AI tool is laundering the lie into the mainstream. A later test by AFP, after the story had been widely debunked, showed that the tools could cite professional fact-checkers and correct themselves. But the inconsistency is itself a problem. Depending on the day, the hour, or the prompt, the same chatbot might either debunk a lie or spread it. Google said it quickly addressed the issue and updated its systems to surface the latest reporting on the topic. Microsoft said the situation was developing rapidly and that once reliable information became available, Copilot reflected the updated context and identified the claim as fabricated. But for a voter scrolling through answers, this kind of whiplash doesn’t inspire confidence. It reminds us that chatbots are not neutral librarians. They are mirrors of the internet’s messiest content, and bad actors know exactly how to exploit them. In the old days, spreading a lie required a platform, a budget, and at least some degree of effort. Now it can happen automatically, with a well-crafted prompt and a naive algorithm.
Consider the larger context. The London-based Institute for Strategic Dialogue pointed out that this election is the first in American history where AI tools are built into every major technology platform. You don’t have to go out of your way to find a chatbot anymore—they pop up in your search engine, your email app, even the place where you buy shoes. The National Conference on Citizenship, a nonpartisan group, conducted a survey and found that two in five Americans have used AI tools to gather information about candidates or ballot measures. That’s a huge number of people who are essentially asking a computer to tell them how to participate in democracy. Yet trust is dangerously low. Only 40 percent of those users trust AI to get basic logistical facts right, like where, when, and how to vote. And a striking 60 percent said they trust it only a little or not at all to be impartial about candidates. So we are in a strange and uncomfortable position: millions of us are using these tools, but most of us don’t really believe them. It’s like stumbling through a dark room with a flashlight that flickers—you keep it on because you have nothing better, but you never let go of the wall. We are guinea pigs in a vast experiment with no informed consent. The technology is being deployed faster than our laws, our ethics, or our common sense can catch up. And the people most likely to be harmed are those who don’t have a backup plan. A voter who asks a quick question, gets a confident answer, and never thinks to double-check is exactly the person a disinformation campaign is trying to reach. In that sense, AI is not just a tool. It is a new front in the battle for the ballot.
Yet there is a surprising twist in this story. Not every finding is bleak. Researchers at the Brennan Center for Justice, a nonpartisan legal and policy institute, decided to test the big chatbots by pushing them hard on election conspiracy theories. They asked about rigged voting machines, stolen elections, and other false narratives. Some might expect AI to be a source of lies, but the center found that most leading chatbots actually pushed back. They refused to validate the conspiracies, even when users challenged them repeatedly. The researchers described the bots as “sympathetic listeners”—they would acknowledge a user’s fear or anger, engage patiently, and still steer them toward the facts. That is a meaningful distinction. It suggests that while AI systems are flawed, they are not inherently corrupt. They can be designed to promote truth and discourage misinformation. In fact, the Brennan Center concluded that chatbots might play a constructive role in fighting election disinformation, especially for people who are skeptical of traditional news sources but willing to have a conversation with a machine. If a voter believes the election was stolen, a chatbot that listens without judgment and then calmly explains why the claim is false could be more persuasive than a news article or a fact-checking website. That is a genuinely hopeful thought. It doesn’t excuse the mistakes, of course. A chatbot that gives you the wrong voting deadline can suppress your vote just as effectively as a malicious troll. But it does mean these tools are not doomed to be part of the problem. They have the potential to be part of the solution—if we are honest about their limits, if we hold tech companies accountable, and if we refuse to treat their answers as holy scripture.
Where does that leave us? In a word: careful. We need to humanize our relationship with AI before it dehumanizes our politics. That means treating chatbots the way we would treat a helpful but slightly unreliable neighbor. You wouldn’t ask a neighbor for voting instructions without checking with the county clerk, and you shouldn’t ask a bot, either. When election season comes around, use AI to generate questions, not final answers. Use it to help you understand the issues, but then go to the official sources—your local election office, the candidates’ own websites, trusted news organizations—and confirm what you’ve learned. And as citizens, we need to demand more from the companies building these tools. They should be transparent about how their models are trained, what data they use, and how they plan to prevent misinformation from spreading. They should be held accountable when their products make mistakes, especially in an election season when a single wrong answer can cost someone their voice in government. This is not just a technical problem; it’s a civic one. In a democracy, information is oxygen. If chatbots are pumping polluted air into our public square, we have a collective responsibility to fix it. The good news is that we still have choices. We can be skeptical, curious, and informed. We can remember that underneath every algorithm are human decisions, human values, and human failures. And we can carry that awareness into the voting booth, where the only voice that really matters is our own.