Here is the content summarized and humanized into 2,000 words, presented in six paragraphs.
—
There is a moment that nearly every voter knows too well: you are standing in the kitchen, scrolling through your phone, trying to figure out where to go to vote, who is on your ballot, or whether that candidate you heard about actually exists. The information is fragmented across government websites, news articles, campaign ads, and social media posts that all seem to contradict each other. So, when a chatbot appears on your screen and promises to answer your questions in a calm, conversational, and instantly organized way, it is almost irresistible. It feels like having a knowledgeable friend sitting at the table with you, someone who has read everything and is happy to break it down. But what if that friend is not actually knowledgeable? What if that friend is confidently making things up? This is the uncomfortable scenario playing out across the United States as the 2026 midterm elections approach. A recent investigation has revealed that AI chatbots, despite their polished and reassuring tone, are frequently giving voters false election information—from imaginary candidates to fundamentally wrong congressional districts. The very tools we turn to for clarity are, in many cases, adding to the fog. It is a troubling reminder that technology, for all its brilliance, is still learning, and when it comes to something as sacred as the right to vote, a made-up answer is more than just an inconvenience; it is a threat to informed democracy.
The Washington Post recently decided to put this issue to the test in a way that was both simple and deeply revealing. They took six popular AI chatbots and asked them to create voter guides for randomly selected addresses across the United States. This is exactly what millions of people might ask a chatbot to do: “Tell me who is running in my district and where I vote.” The researchers collected 3,000 separate responses and analyzed them for accuracy. What they found was not just an occasional slip-up; it was a pattern of concerning misinformation. The chatbots failed to identify the correct congressional district for an address in 48% of their responses. That is nearly half. And even when the chatbots did name a district, they were wrong 22% of the time. Think about the human reality behind those numbers. A voter in Ohio types in their street address, receives a neat list of candidates, and heads to the polls believing they are making an informed choice. But the district they were given belongs to a different part of the city, or the candidate they researched is actually running in a neighboring state. It is not just about getting a fact wrong; it is about breaking a fundamental link between a person and their representation. The study showed that these errors were not random flukes either. They were woven into otherwise articulate, detailed, and seemingly helpful responses, making them even more dangerous because they were hard to spot. The chatbots were not simply saying “I don’t know.” Instead, they were delivering polished, confident, and entirely fabricated information with the same tone used for accurate answers, leaving voters with no easy way to tell the difference.
The details of the chatbot failures are as fascinating as they are alarming. In one instance, ChatGPT invented a Missouri Senate race that was utterly absent from the 2026 ballot. Imagine being a Missouri voter who asks about the Senate election, sees a candidate name, does a quick search, finds nothing, and then wonders if they are the one out of touch. Another chatbot, Claude, presented Wyoming politicians who were running for state-level offices as if they were candidates for federal office. That is a profound misrepresentation of an entire political system. But perhaps the most telling finding was about which chatbots performed “better” in the accuracy rankings. ChatGPT, Google AI Mode, and Perplexity scored higher, but not because they were more intelligent or had better data. They scored higher because they were more cautious. They often refused to provide a specific answer when they were uncertain, choosing to acknowledge their limitations rather than inventing a response. In a world that prizes artificial intelligence for its confidence and fluency, this is a surprising twist: the safest chatbot was often the one that said “I can’t help with that” or “this is not reliable enough to answer.” It turns out that the most humanizing quality these systems can exhibit is humility. When a chatbot admits it does not know something, it is protecting the user from a cascade of misinformation. Refusing to answer is not a failure; in fact, it is a feature. It is a moment of honesty in an otherwise hallucination-prone machine. The fact that we have to praise this as a positive outcome says a lot about how much work still needs to be done to make these tools trustworthy for civic engagement.
One of the most troubling revelations from the investigation is why the chatbots were making so many mistakes. It was not simply a case of broken algorithms or corrupted datasets. Some of the errors traced back to a direct source: other AI-powered websites. The researchers identified a site called The Ballot Brief, which describes itself as an AI-powered election information resource. This site was reportedly using ZIP codes to determine congressional districts. That is a fundamentally flawed approach, because congressional districts are drawn along precise population and geographic lines, and they can split a single city, a single neighborhood, or even a single street. A ZIP code is a postal convenience, not a political boundary. Yet the chatbots were scraping information from this site and presenting it to voters as gospel. This creates a troubling cycle where AI is learning from other AI, and the errors are being laundered through legitimate-looking sources. Even worse, the chatbots often contained biographically accurate information about real candidates, making the fabricated parts harder to detect. A voter might read a long paragraph about a candidate’s background, recognize their name from a yard sign, and assume the rest of the information is equally correct. Because the chatbots also linked to official political websites, they gained an aura of trustworthiness that they did not deserve. The presence of a clean link to a government page at the bottom of a response can fool a person into thinking the entire response was vetted and approved by an official source. In reality, the chatbot was stitching together reliable data with hallucinated details, creating a noisy patchwork of truth and falsehood that would challenge even the most careful reader.
This confusion is not happening in a vacuum. A Pew Research Center survey of 5,273 US adults found that 8% of Americans had already used AI chatbots for information about the 2026 elections. Eight percent may sound like a small slice of the population, but in a country with more than 260 million adults, that translates into millions of people. It is not an obscure fringe behavior anymore; it is a mainstream information habit. The consequences of feeding millions of voters incorrect data are enormous. A chatbot that creates a fake candidate out of thin air could literally sway a vote, because a voter might be convinced to look up that candidate, find nothing, and then question their own research. Or worse, a voter might casually mention the fabricated candidate to a friend, and the falsehood takes on a life of its own. Word of mouth is a powerful thing, especially during election season, and rumors spread faster than corrections ever will. Beyond the immediate risk of misinformed voters, there is a deeper, more insidious damage: the erosion of trust. When people realize that AI chatbots have handed them false information, they may not just lose faith in chatbots. They may lose faith in the electoral process itself. They may conclude that the system is rigged, that no information is real, and that voting is a pointless exercise. A fake candidate does not just confuse one person; it feeds into a broader narrative of corruption and unreliability that undermines democracy. This is precisely why the creators of these chatbots, and the policymakers who oversee them, cannot treat this as a minor glitch to be patched later. It is a civil rights issue, a civic integrity issue, and an urgent one.
So where does this leave the average voter? It would be easy to say that we should all stop using AI chatbots entirely and go back to official sources. But that is neither realistic nor necessary. The larger lesson is that we need to treat chatbots as a starting point, not a final answer. When a chatbot tells you who is on your ballot, you should check that information against your county election office’s official website. When a chatbot tells you that a candidate is running in your district, you should verify the district boundary with a map published by the state or federal government. This is not about being paranoid; it is about being responsible. It is also about teaching the next generation of voters to be curious and skeptical. We engage with technology every day, and we do not have to reject it to protect ourselves. We just have to demand better. We should hold AI companies accountable for the accuracy of their outputs, especially in high-stakes contexts like elections. We should encourage them to program their systems to err on the side of caution, to say “I don’t know” instead of guessing, and to be transparent about their information sources. And we should remember that the human mind is still the most important filter. The kitchen table conversation about who to vote for should still begin with trusted friends, real conversations, and genuine research. A chatbot can be a helpful companion along the way, but it should never be the voice that tells us how to vote. Democracy has always survived because ordinary people care enough to seek the truth. In the age of artificial intelligence, that search for truth will take a little more effort, a little more skepticism, and a lot more wisdom. But it is a search worth taking, because the story of our elections is ultimately the story of who we are as a people.