AI chatbots lean on three obscure, error-prone sites for voting answers

When chatbots answer questions about the US midterms, they often lean on three little-known voter websites, all built with AI. The Washington Post ran tests and found that Microsoft Copilot cited one of them, Ballot Brief, in 75% of its tests. ChatGPT cited the same site in 22%. The Post describes Ballot Brief as error-prone.

What changed

The Post's report points to where chatbots get their information, not just what they say. Three obscure voter sites kept showing up in answers about the midterms. According to the Post, Ballot Brief began with a founder, Ben Torres Ezrick, who was frustrated by how hard it is to find information on political candidates and wanted a one-stop shop for voters.

Who it affects

Anyone who asks Copilot or ChatGPT about candidates, ballots or voting could be shown answers drawn from these sites. That includes voters in the midterms who use a chatbot as a shortcut. A wrong answer about a candidate or a ballot is easy to miss, because chatbots tend to sound sure of themselves.

What is not known yet

The material does not say which specific errors appeared on Ballot Brief, how many were found, or whether the site will fix them. It also does not say whether Microsoft or OpenAI plan any changes. The two percentages come from the Post's own tests, not from a measure of all chatbot use.

What to do

The alert above lists the steps. The short version: treat a chatbot as a starting point, not the final word, and confirm anything about voting with your state or local election office. If a chatbot shows its sources, look at them. A site you have never heard of is a reason to double-check.