Imagine yourself at the kitchen table on a Tuesday evening in October, phone in hand, trying to figure out what will actually be on your ballot. You don’t have time to dig through a county website. You type a question into a chatbot: “Who’s running for Senate in my state?” or “Where do I vote?” The answer appears instantly, polite, specific, and utterly certain. It names candidates. It gives dates. It even includes a brief description of each race. Why would you question it? This is the exact scenario that election officials and researchers have been worried about, and according to The Washington Post, their worry is justified. Testers reviewed 3,000 AI-generated voting guides and found frequent, concrete factual errors. ChatGPT reportedly added a U.S. Senate race to Missouri’s 2026 ballot that was not up for election. It also misidentified Wyoming municipal candidates as contenders for federal office. To be transparent, these particular examples are attributed to Washington Post testing and have not been independently verified from primary sources, but they represent the kinds of errors that would be difficult for most voters to detect without checking official ballot records. That is what makes them so dangerous. A chatbot doesn’t whisper or hedge. It doesn’t say “I might be wrong.” It presents a fabricated race with the same calm authority as a real one, and for a voter who just wants to be prepared, the response looks like the truth. You have no reason to doubt it, because the machine doesn’t sweat. It doesn’t hesitate. It doesn’t offer you a source unless you ask. It simply gives you an answer, and the answer sounds like a knowledgeable friend who has done all the homework for you.
The problem is bigger than any single wrong answer, because misinformation can compound itself before you even ask your question. Chatbot answers may not draw on current official election authorities, and you cannot assume that they do. These systems pull from web content, and that content increasingly includes AI-generated pages that may themselves contain errors. A chatbot can absorb a flawed source, reproduce its claims in confident language, and then that output may eventually circulate as source material elsewhere. Imagine a hall of mirrors: a false statement gets written by one algorithm, scraped by another, repeated by a third, and then returned to you as a definitive answer. Researchers describe this as a plausible and documented risk, though the precise mechanics vary by system and have not been fully measured in publicly available research. But you don’t need a scientific study to feel the effect. Confidence is the specific danger. A response that includes a phantom Senate race alongside accurate candidate biographies gives no obvious signal that anything is wrong. If everything else in the answer checks out, the one false item slips past your attention. It feels authoritative because it looks authoritative. And because chatbots are designed to be helpful, they often include a level of detail that makes their answers feel more reliable, not less. That is the trap. You’re not looking for errors. You’re looking for information. The machine is happy to provide it, even when the information isn’t true. It doesn’t warn you that its sources might be outdated, or that it couldn’t verify the details, or that it is blending together data from different states and different years. It just tells you what you want to hear, in a voice that makes you want to trust it.
Relatively low adoption does not mean low risk. A Pew Research Center survey of 5,273 U.S. adults, conducted October 12–18, 2026, found that 8% reported using AI chatbots for 2026 election information. That figure sounds modest. It stops sounding modest when you consider what those users were actually asking. Among that group, 76% sought candidates’ positions on issues. 71% used chatbots to assess claims they had heard elsewhere. 65% looked up candidates’ backgrounds and careers. And 40% sought logistical information, including registration status, polling places, and mail ballot rules. That last number should concern everyone. Incorrect answers to logistical queries could interfere with a voter’s ability to participate. A wrong answer about a candidate’s stance is a problem for informed decision-making, but a wrong answer about where to vote can stop someone from voting at all. The survey suggests that chatbots are being used as a general-purpose civic assistant, not just as a curiosity. People are trusting these tools with the mechanics of democracy. Eight percent may not sound like a wave, but in a national election, eight percent of voters is millions of people. If even a fraction of that group receives bad logistical information, the consequences are not abstract. They show up in long lines at the wrong polling place, in applications sent to the wrong office, in ballots never counted. The number of users matters less than the nature of the question. A person who asks a chatbot about a candidate’s tax plan can recover from a bad answer by reading more. A person who asks a chatbot where to vote and receives a wrong address may not recover before the polls close.
The address problem is at the heart of why chatbots fail at this task. Election information is local in ways that generalized AI answers are not equipped to handle. Correct voting guidance depends on a voter’s precise address, current district boundaries, registration status, ballot type, and the specific jurisdiction conducting the election. You cannot give a universally correct answer to “Where do I vote?” because the answer changes from street to street. According to Washington Post testing, chatbots failed to correctly identify the congressional district for an exact address 48% of the time. Among responses that attempted a district assignment, 22% were reportedly wrong. That is not a rare edge case. That is a coin flip. And the errors are not always obvious. A plausible answer that names real candidates or real offices could still direct a voter to the wrong polling place, cite the wrong deadline, or describe a race that does not appear on their ballot. The answer may sound perfectly normal. It may include a street address and a time range. It may even mention the correct county. But if the district is wrong, everything downstream is wrong too. The voter shows up on Election Day expecting one ballot and gets another. Or they mail a ballot to an address that no longer accepts it. The chatbot doesn’t know what it doesn’t know. It has no way to tell the difference between a correct answer for one jurisdiction and a correct answer for another. In a system where every voter has a unique combination of location and registration, a one-size-fits-all response is not just incomplete; it is dangerous. Even a perfectly accurate answer for one person can be completely wrong for their neighbor, and the chatbot will never know the difference.
One wrong answer is enough. Misinformation does not need to be widespread to affect participation. Election errors do not need to reach a majority of voters to matter. A single voter with an incorrect polling location or a missed mail-ballot deadline is a voter whose participation could be impaired. The risk is real even when overall chatbot usage remains limited. Think of all the small steps that go into voting: confirming you’re registered, finding your polling place, checking your sample ballot, learning the deadlines, filling out the forms, getting the ballot in on time. A chatbot can provide a wrong answer at any of these steps, and the voter may never know until it’s too late. The anonymity of the interaction makes it worse. When you get bad information from a human, you can ask follow-up questions, sense hesitation, or request a source. A chatbot gives you the same smooth response whether it is certain or guessing. It doesn’t tell you that it used an outdated map or an unreliable webpage. It doesn’t tell you that it made up a race. It just answers. In a way, the very helpfulness of these tools is what makes them risky. The more polished the answer, the less likely you are to verify it. And the cost of that trust can be a lost vote. That is not a hypothetical harm. It is a direct threat to the basic promise that every eligible citizen can participate in choosing their representatives. One wrong answer may not change the outcome of an election, but it can change the life of the voter who acted on it. It can turn a civic duty into a frustrating puzzle, and it can convince someone that the system is broken even when the system was working fine. The error was in the machine, not in the voter, but the voter is the one who pays the price.
Where can you get the right answer? Official sources exist, are free, and are maintained by the authorities who administer each election. For ballot contents, registration status, polling locations, early voting schedules, and mail ballot rules, state and county election authority websites are the appropriate sources. They may not be as conversational as a chatbot, but they are accurate, current, and legally responsible for the information they publish. Chatbots may help a voter understand a candidate’s publicly stated positions or organize information from verified documents, but they are not appropriate as the sole source for any logistical or administrative election question. When a vote depends on getting the details right, the source matters as much as the answer. That doesn’t mean you have to abandon technology. It means you should treat general-purpose chatbots as a starting point, not a final authority. If you use one, ask for the official website it used, then go check that website yourself. If it can’t provide a source, that’s your cue to stop trusting it. The same principle applies to AI-powered voting guides: look for the organization behind them, their methodology, and whether they cite official election records. For those seeking vetted AI-powered websites that can reliably support research, alternatives exist beyond general-purpose chatbots. But the first step is knowing where to look. Your vote is too important to rest on a guess, no matter how confident it sounds. The chatbot will still be there tomorrow, ready to answer your next question. But the deadline won’t wait. The polling place won’t move. The ballot won’t change. Getting the right answer from the right source is not just a matter of convenience; it is the foundation of making your voice count.