Copilot cites error-prone Ballot Brief in three-quarters of US midterm tests
Washington Post tests show Copilot and ChatGPT frequently reference obscure, AI-built voter information sites.
Key points
- Copilot cited Ballot Brief in 75 per cent of Washington Post tests.
- ChatGPT cited the same site in 22 per cent of those tests.
- The sites were built by Ben Torres Ezrick to consolidate political data.
according to the published report, Microsoft Copilot cited the error-prone Ballot Brief site in 75 per cent of its tests regarding US midterm election information.reported OpenAI’s ChatGPT referenced the same obscure source in 22 per cent of the tests conducted by the newspaper.reportedTests show Copilot and ChatGPT often use a single, error-prone site for election answers.Washington Post tests indicate significant reliance on Ballot Brief by Copilot and ChatGPT.
Evidence: reported
Tests: chatbots rely on three obscure AI-built voter sites for US midterm answers; Copilot cited the error-prone Ballot Brief in 75% of tests and ChatGPT in 22% (Washington Post)aifleet-techmeme
Evidence: reported
Tests: chatbots rely on three obscure AI-built voter sites for US midterm answers; Copilot cited the error-prone Ballot Brief in 75% of tests and ChatGPT in 22% (Washington Post)aifleet-techmeme
How we got here
Ben Torres Ezrick built the voter information sites after becoming frustrated with the difficulty of finding political candidate details. He aimed to create a centralised resource for voters seeking information on candidates running in the US midterm elections.
What is new
Testing revealed that major AI assistants cited this specific, AI-built source in a significant portion of queries about midterm elections. The Washington Post found that Copilot cited Ballot Brief in 75 per cent of tests, while ChatGPT cited it in 22 per cent.reported
Evidence: reported
Tests: chatbots rely on three obscure AI-built voter sites for US midterm answers; Copilot cited the error-prone Ballot Brief in 75% of tests and ChatGPT in 22% (Washington Post)aifleet-techmeme
What to watch
Developers should monitor how often their models cite niche or unverified sources for high-stakes civic information. The reliance on error-prone sites indicates a need for better filtering mechanisms in AI search implementations.
Availability and limits
Users accessing Copilot or ChatGPT for US election data may receive answers derived from Ballot Brief. The Washington Post documented that these citations occurred in a significant portion of their specific test cases. No independent verification of all site content is provided by the assistants.
Why it matters
- AI application developers must consider how their models select sources for civic queries.
- The high citation rate of obscure sites highlights potential reliability issues in current AI search results.
- Users seeking election information may encounter errors if assistants cite these specific platforms.
- The findings suggest a need for improved source validation in AI systems handling political data.
Topics: AI assistants, source verification, US elections, Copilot, ChatGPT
Updated
Then / now / next
Users struggled to find consolidated information on political candidates.
AI assistants frequently cite a specific, error-prone site built by one developer.
Developers may need to adjust source selection strategies for civic information.
Story timeline
- Then Users struggled to find consolidated information on political candidates.
- 01 Nov 2026 Tests: chatbots rely on three obscure AI-built voter sites for US midterm answers; Copilot cited the error-prone Ballot Brief in 75% of tests and ChatGPT in 22% (Washington Post)
- Now AI assistants frequently cite a specific, error-prone site built by one developer.
- Next Developers may need to adjust source selection strategies for civic information.
The strongest doubt
Counterpoint from the sources
The site was built to simplify voter information access, though it contains errors.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
Ballot Brief
An obscure, AI-built voter information site.
Midterm elections
US elections held halfway through a presidential term.
Questions readers ask
Which AI models cited Ballot Brief?
Microsoft Copilot cited it in 75 per cent of tests, and ChatGPT in 22 per cent.
Who built Ballot Brief?
Ben Torres Ezrick built the site to consolidate voter information.
What was the source of the tests?
The Washington Post conducted the tests on US midterm election answers.
Connected stories
Reading receipt
- Copilot cited Ballot Brief in 75 per cent of Washington Post tests.
- ChatGPT cited the same site in 22 per cent of those tests.
- The sites were built by Ben Torres Ezrick to consolidate political data.
Still open: How many other obscure sites are cited by AI assistants for civic data?
Still open: Will vendors adjust source selection algorithms following these findings?