Voters looking for help with candidates and ballot questions may be tempted to ask a chatbot, but new research suggests that is a poor idea. The evidence against turning to chatbots to use AI to vote is mounting: major systems including ChatGPT, Claude, and Gemini have been shown to produce outdated, mistaken, or politically slanted answers about elections.
A habit already widespread
Chatbot use is no longer a niche practice. Pew Research found that roughly half of surveyed US adults have used an AI chatbot at some point, and 8 percent said they had used such tools ahead of the midterms to look up information on candidates and policies.
The problem is not that people are asking — it is what they are getting back.
Errors are common, even on basic facts
Research by Forum AI found that nearly 40 percent of election-related answers from mainstream chatbots contained at least one factual error. Some mistakes were glaring. Google’s AI Mode declined to give the results of the 2020 elections, and the company’s Gemini chatbot stated that Pennsylvania’s deadline to request a mail-in ballot was November 3, when the actual deadline was October 27.
Other failures were more subtle. When Forum AI researchers asked in August why someone should vote for North Carolina congressman Don Davis, Claude’s response merged the lawmaker with a former school superintendent who shares his name.
Google and Anthropic did not respond to requests for comment. OpenAI pointed to a blog post describing its approach to elections in 2026.
The sourcing problem behind the errors
Part of the difficulty is where these systems draw their information. Chatbots depend on a pool of online content that is increasingly compromised. NewsGuard research found that mainstream chatbots cited so-called “pink slime” websites — outlets that look like ordinary news organizations but are funded by political or ideological groups and individuals — nearly 50 percent of the time when asked about candidates and policies in the midterms.
Those sites have multiplied ahead of the elections, and chatbots are picking them up. ChatGPT was the worst offender in NewsGuard’s analysis, citing pink slime sources in more than 70 percent of searches. xAI’s Grok performed best, citing them in 30 percent of cases.
Overpromised and under-sourced
Nina Jankowicz, executive director of the American Sunlight Project and a former disinformation official during the Biden administration, told WIRED that most Americans are not equipped to navigate an election in the age of consumer AI. She pointed to companies having overpromised what the technology can do. Because chatbots must ingest information from somewhere, and those sources are often weak, the systems frequently invent answers when they do not know one.
The result is a tool that can sound confident while being wrong — a dangerous combination when the subject is how to cast a ballot or which candidate to support.
What voters should keep in mind
- Treat chatbot answers about elections as a starting point for verification, not a final authority.
- Check deadlines and voting rules with official election offices rather than relying on a generated response.
- Be aware that citations and links from chatbots may point to partisan or disguised sources.
- Remember that a confident tone is not evidence of accuracy.
For an election decision, the safer path is to consult official information directly. The research suggests that, at least for now, using a chatbot as a voting guide carries real risk of being misled.
