How did Ballot Brief, a site almost no one visits, become an election information source for ChatGPT and Copilot?

A Washington Post test showed that ChatGPT cited Ballot Brief in 22% of election-related answers, and Microsoft Copilot did so in 75% of them. But the site's candidate profiles contain multiple errors, and one of those errors was repeated by Copilot to a questioner.

Ezik, who previously held senior positions at Google, Zoom and the video platform Rumble, found that when voters want to learn about candidates, they often have to spend considerable time searching across different websites, so he decided to build a one-stop site providing election information.

Faced with profiles for thousands of candidates, he chose to let AI help. He dispatched multiple groups of digital agents to gather information online, then placed the organized content into Ballot Brief for people to consult.

The site attracted far more than human visitors. Ezik said his data showed that over the 30 days spanning September and October, Microsoft's AI systems cited Ballot Brief more than 680,000 times.

The Washington Post investigation found that Ballot Brief is just one of them. Three new websites, including it, frequently appear among the voting information sources provided by chatbots. The founders of all three sites said their teams are very small and rely almost entirely on AI to collect and summarize data.

The problems followed. Isabel Linzer, an elections analyst at the Center for Democracy and Technology, believes the findings show that information one person publishes on an AI-generated website can easily be adopted by chatbots. She said this could affect voters' trust in election information.

The Washington Post found multiple errors on Ballot Brief.

One profile on the site said that Ohio Republican Senator Jon Husted supported a bill. But that bill had already passed before he took office. Copilot repeated this error when answering The Washington Post's question.

The site also merged two California candidates both named Robert Garcia into a single profile, and the committee assignments of several other House members were also listed incorrectly.

Some lawmakers' profiles were quite brief, lacking voting records, and the site also incorrectly noted that they appeared to have no campaign website.

Ezik said he works hard to ensure the information is accurate. After The Washington Post showed him the findings, he made several changes to how the site operates.

He said the site's purpose is to provide voters with as much information as possible, but it should not be their only source of information.

A test covering 3,100 counties

To understand how chatbots answer election questions, The Washington Post partnered with the data company Promptwatch to test AI products from OpenAI, Anthropic, Google, Microsoft, Amazon and Perplexity.

For about 3,100 counties in the United States, researchers selected one address in each county and asked the chatbots the same question: to create a November 2026 voter guide based on that address, listing all elections and the candidates' positions on major issues.

As a result, the chatbots often failed to correctly determine which congressional district the address belonged to, and sometimes simply refused to answer.

The citation percentages above come from this specific question and test environment, and do not mean that ChatGPT or Copilot will cite Ballot Brief at the same rate in all election questions.

The Washington Post found that in the first round of testing in September, Ballot Brief was the most frequently cited source when ChatGPT answered such questions. After the newspaper informed OpenAI of the results last week, in the second round of testing, the proportion of ChatGPT citing the site dropped to 4%.

OpenAI said the company hopes to help people participate in elections through reliable, authoritative information sources and measures to guard against bias and abuse; when problems are pointed out, the company takes action to improve model answers. The Washington Post disclosed in its report that it has a content partnership with OpenAI.

Ballot Brief was also the most frequently cited source for Copilot in the tests. After the report was published, Microsoft said it had adjusted the tool to prioritize authoritative sources and official state election websites, and would continue testing and improving election-related answers.

Anthropic's Claude mainly cited established voter guides run by nonprofit organizations in the first round of testing, but in the second round it began frequently linking to Ballot Brief. Anthropic did not explain the change, saying only that the company had tested the model before the midterm elections, strengthened safeguards, and worked to provide reliable and accurate voting information.

Mowafak Allaham, a doctoral researcher at Northwestern University, found in a study that for questions in areas such as politics, health and the environment, **16%** of the sources cited by chatbots were themselves AI-generated.

He said it is not transparent how these systems find, filter and integrate information.

Two other researchers who previously worked on data analysis for Kamala Harris's 2024 presidential campaign found that chatbots also relied heavily on AI-generated websites when answering midterm election questions. Their tests also found that Ballotpedia, a long-standing election information site, had been blocking ChatGPT.

Ballotpedia CEO Leslie Graves said the blocking was caused by an error in a security setting, and the site lifted the restriction last week. Ballotpedia has a team of about 90 people, while Ezik and others hope to use AI to build a huge candidate database on a smaller scale.

In addition to Ballot Brief, The Washington Post's tests also found that chatbots frequently cited DecodeTheVote and Smarter.Vote. The former uses AI characters to introduce candidates' backgrounds, and the latter also attempts to predict election results.

Jacob Loukota, founder of Smarter.Vote, said his site has been cited by AI bots 1.7 million times, far exceeding visits by human visitors. He believes chatbots may like citing the site because it indicates the sources of its information.