Here's Why You Shouldn't Use AI to Vote

Pew data cited by WIRED says roughly half of US adults have used an AI chatbot, and 8 percent say they used one to research candidates and policies before these midterms. That's a small share, but it's a lot of people, and the share will only grow.

The error rate is the problem. Forum AI found that almost 40 percent of election answers from mainstream chatbots had at least one factual error. One of them was Gemini giving the wrong Pennsylvania mail-ballot request deadline. NewsGuard found chatbots citing partisan "pink slime" sites, outlets that look like local news but are funded by political interests, in nearly half of midterm queries.

A wrong deadline doesn't just misinform. It can cost someone a ballot.

We use these tools every day and we think they're genuinely useful for plenty of research. Elections are a different category. The facts are local, they change on a schedule, and getting one detail wrong has consequences you can't undo after the polls close. That's exactly where today's chatbots are weakest.

The findings here point to two separate failures. One is the familiar one: the model doesn't know and fills the gap with something that sounds right, like Claude confusing a sitting congressman with a former school superintendent of the same name. The other is quieter and, to us, more worrying. When the bots do search the web, they're pulling from an information pool that political operators have learned to seed. NewsGuard found ChatGPT citing pink slime sites in more than 70 percent of searches. Grok did best of the group and still cited them in 30 percent.

That second failure scales badly. Every new partisan site dressed up as local news becomes a potential source for millions of AI answers. The incentive for operatives is obvious, and the cost of building such sites keeps falling. If AI search becomes a default way people learn about candidates, the fight over what the models cite becomes a campaign front.

The responsibility sits with the companies first. Google and Anthropic didn't comment to WIRED; OpenAI pointed to a blog post about its election approach. We'd like to see more: hard routing to official election offices for deadlines and polling places, clear labels on who funds the sources being cited, and published accuracy testing before every major election, not just reactions to outside audits.

Until then, our advice is plain. Use a chatbot to understand an issue if it helps. Get dates, rules and candidate facts from your state or county election office.

Forum AI and NewsGuard ran independent tests of mainstream chatbots ahead of the 2026 midterms. Nina Jankowicz of the American Sunlight Project told WIRED that companies have overpromised what AI can do and that models often make things up when they don't know the answer.