Microsoft's Copilot and OpenAI's ChatGPT frequently source U.S. midterm election information from obscure, AI-built voter sites, according to recent tests. Copilot cited the error-prone Ballot Brief in 75 percent of queries, while ChatGPT used the same platform in 22 percent of its responses.
Ballot Brief has been identified as error-prone, meaning users seeking election information from these chatbots may receive incorrect data. The reliance on a limited set of less authoritative sources introduces significant accuracy risks, particularly on sensitive civic topics where misinformation carries real consequences.
Differences in citation rates between Copilot and ChatGPT suggest variations in their underlying data retrieval and ranking algorithms. Copilot's higher dependency on these specific, less reliable sources indicates weaker source prioritization logic.
The economics here matter. User confidence underpins both consumer adoption and enterprise deployment. Consistent accuracy failures in high-stakes domains like elections risk eroding trust and limiting TAM expansion in institutional markets. For AI developers, this is a competitive vulnerability: source verification and data diversification become table stakes for market viability.
