AI Chatbots Give Voters Unreliable Election Information

About half of U.S. adults surveyed said they had used an AI chatbot, while 8 percent said they had used one to research candidates or policies ahead of the midterms. Researchers found that such tools can give voters inaccurate information, even when asked basic election questions.

A Forum AI review found that nearly 40 percent of election answers from mainstream chatbots had at least one factual error. Examples included Google’s AI Mode declining to provide the 2020 election results, and Gemini giving the wrong Pennsylvania mail-ballot request deadline. In another test, Claude mixed up North Carolina Rep. Don Davis with a former school superintendent who shared his name.

The quality of chatbots’ answers is also tied to the websites they draw on. NewsGuard found that mainstream chatbots cited “pink slime” sites—websites that resemble news outlets but are funded by political or ideological groups—in nearly half of searches about midterm candidates and policies. ChatGPT cited them in more than 70 percent of cases, while Grok did so in 30 percent.

Nina Jankowicz of the American Sunlight Project said companies had overstated what AI can do and warned that chatbots may invent answers when they lack reliable information. Google and Anthropic offered no reply to inquiries seeking comment. OpenAI referred questions to a blog post about its approach to the 2026 elections.

Faulty AI Election Answers Put Voters at Risk

AI companies are offering election information through chatbots that can mislead voters at a consequential moment. On November 2, as Americans went to the polls, research showed that these systems could supply outdated details, confuse candidates and cite websites funded by political or ideological interests.

A Forum AI review found factual errors in nearly 40 percent of election answers from mainstream chatbots. Gemini gave the wrong Pennsylvania mail-ballot deadline, while Claude confused Rep. Don Davis with a past school district leader bearing an identical name. Google’s AI Mode also declined to provide the 2020 election results. These are not minor flaws when people are relying on automated tools to understand candidates and voting.

NewsGuard found that chatbots cited “pink slime” sites in nearly half of searches about midterm candidates and policies. ChatGPT did so in more than 70 percent of cases. Nina Jankowicz of the American Sunlight Project warned that companies had overstated AI’s capabilities and that chatbots can make up answers when information is missing. Google and Anthropic gave no comment when approached; OpenAI referred questions to its elections blog. Voters deserve reliable information, not systems that can amplify partisan material while presenting answers with confidence.

Voters Should Not Trust AI Chatbots for Election Facts

The central lesson for voters on November 2 is straightforward: an AI chatbot is not a dependable substitute for checking election facts. Research found that mainstream systems can return incorrect deadlines, muddle candidates and draw on sites funded by political or ideological groups.

Forum AI found at least one factual error in nearly 40 percent of the election answers it assessed. Gemini gave the wrong deadline for requesting a Pennsylvania mail ballot, and Claude confused Rep. Don Davis with a namesake who was a former school superintendent. Google’s AI Mode declined to provide the 2020 election results. These failures show why voters should be cautious about treating chatbot responses as authoritative.

NewsGuard found that chatbot searches about midterm candidates and policies cited “pink slime” sites nearly half the time. ChatGPT’s rate exceeded 70 percent, while Grok’s was 30 percent. Nina Jankowicz of the American Sunlight Project said AI companies had overpromised and noted that chatbots can invent responses when they lack an answer. Google and Anthropic offered no reply to inquiries seeking comment, and OpenAI pointed to a blog about its 2026 election approach. The findings make clear that convenient technology cannot guarantee accurate civic information.