As Americans head toward the 2026 midterm elections, some are turning to artificial-intelligence chatbots for information about candidates and policies. But research and recent examples suggest that even supposedly nonpartisan systems can provide outdated or fabricated information, make factual errors, and rely on politically funded websites.
About 50 percent of U.S. adults recently surveyed said they had used an AI chatbot at least once. According to Pew Research, 8 percent said they had used these tools to find information about candidates and policies ahead of the midterm elections.
However, much of the information produced by chatbots is outdated, hallucinatory or drawn from partisan websites. OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini have all proved inadequate when answering questions about the midterm elections.
Errors in election information
Just last week, Google’s AI Mode refused to provide the results of the 2020 election. Google’s Gemini chatbot also said that the deadline to apply for a mail-in ballot in Pennsylvania was November 3, rather than the actual deadline of October 27, according to a Forum AI survey.
According to Forum AI research, almost 40 percent of election-related responses from mainstream chatbots contained at least one factual error. In August, when researchers asked Claude why they should vote for North Carolina Congressman Don Davis, the chatbot confused him with a “former school superintendent” who has the same name.
Google and Anthropic did not respond to requests for comment. OpenAI shared with WIRED a blog post outlining its approach to election safeguards ahead of the 2026 election.
Chatbots and partisan websites
Part of the problem is that chatbots rely on increasingly unreliable online information pools. An investigation by NewsGuard found that mainstream chatbots cited so-called “pink slime” websites nearly 50 percent of the time when asked about candidates or policies related to the midterm elections.
Although these sites present themselves as local or mainstream news outlets, they are funded by political or ideological groups or individuals. More such sites are appearing ahead of the midterm elections as chatbots gain traction.
NewsGuard identified OpenAI’s ChatGPT as the worst offender, citing “pink slime” sites in more than 70 percent of searches. xAI’s Grok performed best among the chatbots examined, citing those sites in 30 percent of searches.
“I don’t think most Americans are ready to navigate the election in the age of consumer AI,” Nina Jankowitz, executive director of the American Sunlight Project and a former disinformation czar in the Biden administration, told WIRED.
“One of the problems is that executives and companies are actually over-promising about AI’s capabilities. AI has to get this information from somewhere, and often that source is not very good. And because of its predictive capabilities, AI often makes up stories even when it doesn’t know the answer.”
Source: www.wired.com
