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

Even the largest ostensibly non-partisan systems have stumbled on basic election facts. Forum AI found Google's AI Mode declining to give 2020 election results, and Gemini misstating Pennsylvania's mail-in ballot request deadline as November 3 when the actual date was October 27. Across Forum AI's testing, close to 40 percent of election answers from mainstream chatbots carried at least one factual error. In an August probe, Claude was asked why someone should vote for North Carolina congressman Don Davis and answered with a description of a former school superintendent sharing the same name.

A structural issue sits underneath the mistakes: chatbots are drawing on a degraded information pool. NewsGuard research found that mainstream chatbots cite 'pink slime' sites — outlets that look like conventional news but are funded by political or ideological interests — nearly half the time when asked about midterm candidates and policies. Such sites have proliferated ahead of the midterms. OpenAI's ChatGPT was the worst performer, citing them in more than 70 percent of searches, while xAI's Grok was the strongest at 30 percent.

Google and Anthropic did not answer requests for comment; OpenAI pointed WIRED to a blog describing its elections approach for 2026. Nina Jankowicz, executive director of the American Sunlight Project and a former Biden-administration disinformation official, argued the industry has oversold AI's abilities, noting the systems must ingest information from somewhere, often unreliable sources, and that a predictive model will fabricate when it lacks an answer.

What remains open is whether vendors tighten retrieval and citation standards before the next cycle, how researchers and platforms treat the growing pink-slime footprint, and whether voters themselves adjust how much weight they give chatbot answers on Election Day — questions that will shape the reliability of AI as a civic information layer.

For now, the concrete takeaway is that error rates near 40 percent and citation rates above 70 percent for ideologically funded sites make chatbots a weak primary source for voting decisions, pushing verification back onto the voter.