ChatGPT, Gemini become ad hoc voter guides in 2026 midterms
TL;DR
- A September survey by the American Association of Political Consultants Foundation found one in five voters consult AI chatbots for election guidance, at similar rates across parties.
- NPR profiled voters using ChatGPT, Gemini and Claude to compare governor candidates, research down-ballot races with thin coverage, and verify viral claims.
- Johns Hopkins postdoctoral fellow Rafael Batista warns chatbots may reinforce the views users already hold, leaving them more confident but not better informed.
About one in five voters now consult AI chatbots for election news and voting guidance, with Democrats and Republicans using them at similar rates, according to a September survey by the American Association of Political Consultants Foundation. NPR profiled voters who leaned on ChatGPT, Gemini and Claude to work through their 2026 midterm ballots.
Adam Johnson, a 40-year-old graphic designer in Morgantown, West Virginia, spent hours walking ChatGPT through his political principles before quizzing it on Senate and county races. "ChatGPT is just kind of a people pleaser," he told NPR, though he added, "I wouldn't feel too bad if I had to do the election today." Lisa Veldran, 65, a retired Madison City Council staffer, used Gemini to build a comparison table of gubernatorial candidates after one dropped out of the race. In Exton, Pennsylvania, Lakshmi Iyer, 50, built a civic site called Ballot Lookup on top of Claude, instructing the model to draw only from candidates' own statements and official sources.
Rafael Batista, a postdoctoral fellow studying AI at Johns Hopkins University, cautions that the same tools can serve a user's priors back to them. Chatbots may "reinforce and persuade you even more towards the way that you were leaning already," he said. The warning lands against baselines Pew has already measured: roughly 42% of chatbot users turn to them for information, and about half of American adults use chatbots at all. OpenAI told NPR it "monitor[s] bias in our models to keep ChatGPT's responses politically neutral," though AI companies broadly do not disclose their training sources.
An August study from the States United Democracy Center found that "ChatGPT and Google AI offer incomplete and inconsistent responses to common voter questions, even as the accuracy of AI platforms improves." Two of the AI researchers we track picked up the NPR piece on the day it ran.
Shared on Bluesky by 2 AI experts
-
No, make it stop People are asking ChatGPT to help them decide how to vote in the midterms www.npr.org/2026/10/05/n...
View on Bluesky →
Originally reported by npr.org
Read the original article →Original headline: People are asking ChatGPT to help them decide how to vote in the midterms