On election day 2026, the chatbot has become one of the first places Americans go to ask about voting — and the numbers published on Tuesday do not make comfortable reading for the companies that built it.
Pew Research found that around 50 percent of surveyed US adults have used an AI chatbot at some point, and 8 percent say they turned to these tools ahead of the midterms to find information on candidates and policies. Research by Forum AI found that almost 40 percent of election answers given by the mainstream chatbots contained at least one factual error. Last week, Google's AI Mode refused to provide the results of the 2020 election, and Google's Gemini said the deadline to request a mail-in ballot in Pennsylvania was 3 November — the real deadline was 27 October. In August, when Forum AI researchers asked why someone should vote for North Carolina congressman Don Davis, Claude confused the lawmaker with a former school superintendent of the same name.
Part of the problem is the well the models draw from. NewsGuard found that mainstream chatbots cite so-called "pink slime" websites nearly 50 percent of the time when asked about candidates and policies related to the midterms. Pink slime sites present as ordinary local news outlets but are funded by political or ideological groups and individuals. Many more of them have appeared ahead of these elections. The worst offender, according to NewsGuard, was OpenAI's ChatGPT, which cited such sites in more than 70 percent of searches; xAI's Grok performed best, at 30 percent.
"I don't think that most Americans are prepared to navigate an election in the age of consumer AI," Nina Jankowicz, executive director of the American Sunlight Project and a former disinformation official in the Biden administration, told WIRED. "One of the problems is that executives and companies have really overpromised on what AI is capable of. It has to ingest this information from somewhere, and often those sources are not very good. And it's predictive, so if it doesn't know the answer, it does frequently make things up."
Google and Anthropic did not respond to WIRED's requests for comment; OpenAI pointed to a blog post outlining its approach to the 2026 elections. The structural point stands regardless: a system tuned to produce fluent, confident prose is being asked for a single verifiable fact — a date, a district, a candidate's record — in the one context where a plausible wrong answer costs the most. A 40 percent error rate is not a rounding problem when the question is where and when to vote.
The practical advice is unglamorous and worth restating on the day itself: state and county election offices publish deadlines and polling places, and they are accountable for them. A chatbot is not.
