AI chatbots are giving voters false and misleading information as the United States heads toward the Nov. 3 midterm vote. The errors range from invented races to candidates placed in the wrong elections, all delivered with the confident tone people often expect from tools such as ChatGPT, Copilot, and Gemini.
The problem is not limited to one strange answer. A Washington Post test found that chatbots struggled to identify basic details about where people vote and which candidates belong on their ballots. The results show how a tool that can explain political issues can also send voters toward a race that does not exist or a candidate who is not running for the office named.
Voter guides filled with basic errors
The Washington Post challenged six of the most popular chatbots to create voter guides for addresses in every county in the United States. The test covered 3,000 AI-generated voting guides and examined the races and candidates relevant to those addresses.
In one test, ChatGPT created a voter guide for a Missouri resident and added a Senate seat that was not up for re-election. In Wyoming, a chatbot described local politicians running in municipal elections as candidates for federal seats. The investigation also found a House candidate in Washington who was not on the ballot.
The location problem was just as striking. Chatbots failed to identify the congressional district for an exact address 48 percent of the time. When they did identify a district, they gave the wrong answer 22 percent of the time.
That leaves voters with a nearly 48-percent error rate before they even reach the question of what each candidate supports. The guides also repeated false information that often traced back to other AI-generated websites, creating a loop in which one unreliable system fed another.
AI-powered websites supplied questionable information to chatbots, and those chatbots then presented the material as voting help. The result was a chain of errors that could make false election information look polished and trustworthy.
A small user base with a large information problem
A late-October survey by the Pew Research Center found that just eight percent of U.S. voters had used an AI chatbot like ChatGPT, Copilot, or Gemini to look up information relevant to the ballot box. That share is still small, but the Nov. 3 vote marks the first federal election in which millions of voters are relying on AI tools for political information.
Researchers say these systems are helping shape the information ecosystem around the polls while also adding to the confusion. Chatbots can debunk conspiracies and educate voters, but multiple studies show that they often draw from unreliable material on the open web. That behavior erodes public trust and adds more pollution to an internet already filled with election misinformation.
The threats now include deepfake videos and AI-powered search results that spread incorrect information. Chatbots have also been accused of regurgitating Russian propaganda, adding another layer to the information chaos surrounding the U.S. midterm elections.
Google faced a separate warning sign last week when its AI Mode would not answer a question about who won the 2020 election. Google said it was working to fix the problem, but the incident showed how AI systems can fail even on major political facts.
Companies prepare election “war rooms”
OpenAI and Anthropic will each have a “war room” dedicated to improving the accuracy of election information during the midterms. The move marks a first for both companies and reflects the need to handle problems that standard product updates may not catch.
The idea of a technology platform “war room” dates back to 2018, when Meta created one amid concerns about election-related misinformation. The AI age has added new risks, including deepfakes, AI-generated websites, and search results that combine incorrect material into answers that sound certain.
An Anthropic spokesperson said, “Teams across the company work year-round on strengthening our election-related safeguards and enforcing our policies. Ahead of the election, we have a dedicated war room monitoring our platform for misuse and ready to quickly respond to any incidents.”
Former Meta director of public policy Katie Harbath described war rooms as “built for dealing with the things we can’t necessarily predict going wrong and you need a quick, established process in terms of triaging.” In plain terms, the teams exist to spot unexpected failures and decide how to respond before those failures spread.
The challenge for voters remains simple but serious: an AI-generated voter guide can look complete while containing an invented race, a misidentified candidate, or the wrong congressional district. The tools may offer useful explanations, but the Washington Post tests show that confident wording is no guarantee that the ballot information is correct.
Based on
- ChatGPT Is Giving Voters Ludicrous Misinformation Right as the Midterms Hit — futurism.com
- Spewing U.S. election falsehoods, AI chatbots face scrutiny – The Japan Times — japantimes.co.jp
- We asked six chatbots to make voter guides. They made serious errors. – The Washington Post — washingtonpost.com
- Election ‘war rooms’ enter the AI age | Semafor — semafor.com
