Copilot Cited Error-Prone Ballot Brief in 75% of Tests — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Chatbots answering US midterm questions relied on just three obscure AI-built voter-information sites, narrowing the source pool behind automated candidate guidance.
- Copilot cited Ballot Brief in 75% of tests, versus 22% for ChatGPT, and Washington Post describes that voter-information site as error-prone.
- Ben Torres Ezrick built a one-stop shop for political candidate information out of frustration with hunting for candidate details — the human origin behind the kind of AI-built voter sites the chatbots turned to.
Why it matters: Ballot Brief, the site the Post calls error-prone, supplied 75% of Copilot's citations and 22% of ChatGPT's during midterm testing. Because the chatbots drew on only three obscure AI-built voter sites, the information behind candidate answers rested with a narrow set of little-known platforms, one of them flagged for errors.
Ask SkimNews
