Chatbots Cited AI-Built Election Site in Up to 75% of Answers, Tests Show

Chatbots pointed voters to a small, AI-assembled election website in a majority of tested queries, according to a Washington Post review of thousands of artificial intelligence responses. Microsoft's Copilot cited the Ballot Brief in 75% of its answers to election questions, while ChatGPT cited it in 22% of answers. Ben Torres Ezrick, who founded the Ballot Brief, said his own data shows Microsoft's AI systems referred to the site more than 680,000 times during a 30-day period spanning September and October.

The Ballot Brief is one of three small websites that chatbots including ChatGPT regularly turn to for voting information, the Post's review found. The other two are DecodeTheVote, which uses AI personas to describe candidates' records, and Smarter.Vote, which also forecasts race outcomes. Jacob Loukota, Smarter.Vote's founder, said his site has been cited 1.7 million times by AI bots, a figure that far exceeds his human visitor count. He said chatbots appear to favor the site because it cites its sources. "That wasn't something I planned, but it makes sense," Loukota said. "It ends up being a natural fit for an AI chatbot."

Ezrick, who previously held senior roles at Google, Zoom and Rumble, said he built the Ballot Brief after struggling to find candidate information in one place. He sent teams of digital agents across the web to assemble dossiers on thousands of candidates. The site is designed to serve both people and chatbots, he said. "It's about providing as much information to the voter as possible but not being their sole resource," Ezrick said.

The Post identified errors in the Ballot Brief's records. Sen. Jon Husted, an Ohio Republican, was credited with backing a bill that passed before he took office, an error Copilot repeated. Two candidates named Robert Garcia running in California were merged into a single profile, and several House members were listed with incorrect committee assignments. Some congressional profiles lacked voting records and carried a note incorrectly stating that no campaign website could be found. Ezrick said he made changes to how the site works after the Post shared its findings.

The Post worked with the data firm Promptwatch to question chatbots made by OpenAI, Anthropic, Google, Microsoft, Amazon and Perplexity. Each was asked to build a voter guide for November 2026 for addresses tied to each of the nation's 3,100 counties. The bots often placed addresses in the wrong congressional district or declined to answer. ChatGPT relied on the Ballot Brief as its top source in September; after the Post shared results with OpenAI last week, that figure fell to 4% in a second round of tests. Microsoft said after publication that Copilot now prioritizes authoritative sources and official state election sites. Anthropic said it had taken steps ahead of the midterms to strengthen safeguards and provide reliable voting information.

Isabel Linzer, an elections analyst at the Center for Democracy and Technology, said the findings show how easily chatbots can come to rely on information "some random person publishes on an AI-generated website." She added, "It's really important to think about how that could affect voter trust." Mowafak Allaham, in doctoral studies at Northwestern University, found that 16% of sources chatbots cited across politics, health and environment questions were themselves made by AI. "These systems are not transparent when it comes to how information is being sourced and curated and synthesized," Allaham said.

Ballotpedia, a long-established site, was blocking ChatGPT, a pair of researchers who led analytics for Kamala Harris's 2024 presidential campaign found. Leslie Graves, Ballotpedia's chief executive, said faulty security settings caused the block and that it was lifted last week. Ballotpedia relies on a team of roughly 90 people.