Technology

Studies find AI chatbots frequently err on election information

Research cited by NPR and Wired flags voting errors and partisan sources in chatbot answers ahead of midterms.

Published Nov 3, 2026 · 23:47 UTC2 reports

The short version

Research cited by NPR and Wired finds popular AI chatbots giving flawed answers on voting and candidates ahead of the midterms.[NPR · Wired]
In IFPG data from October, 13.9% of chatbot answers to voting-logistics questions were outdated or inaccurate.[NPR]
Forum AI research found almost 40% of election answers from mainstream chatbots contained at least one factual error.[Wired]
NewsGuard found mainstream chatbots cite partisan 'pink slime' sites nearly 50% of the time, with ChatGPT highest and Grok lowest.[Wired]

IFPG data from October found 13.9% of chatbot answers on voting logistics were outdated or inaccurate.[NPR]
According to IFPG, Claude and Copilot produced three-quarters of the clear errors found in its testing.[NPR]
Forum AI research found almost 40% of election answers from mainstream chatbots contained at least one factual error.[Wired]
NewsGuard found mainstream chatbots cite 'pink slime' sites nearly 50% of the time when asked about midterm candidates and policies.[Wired]
According to NewsGuard, ChatGPT cited 'pink slime' sites in over 70% of searches, while Grok cited them in 30% of cases.[Wired]

Error rates come from specific test questions and differ across studies, so how well they represent all voters' queries is not established.[NPR · Wired]

Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.