A Washington Post investigation published October 30 found that AI chatbot voter guides often pointed readers to the same obscure source, and that source had documented district errors. Microsoft Copilot and ChatGPT both cited Ballot Brief, a largely AI-produced website, in their election answers. With the midterms on Tuesday, November 3, the findings reach voters with little time to act.
How AI chatbot voter guides went wrong
The Post sent its test requests on September 23, 2026, asking six popular chatbots to build voter guides for addresses in every US county, according to a syndicated reprint of the report. Its tests covered more than 3,000 AI-generated voter guides. The core failure was basic: the chatbots often could not say which congressional district an address falls in, even though every voter has a House race on the midterm ballot.
In 48% of responses, the chatbots did not state a congressional district at all. Guides that did name one got it wrong 22% of the time. A wrong district is more than a typo, because it can lead a voter to weigh candidates they cannot vote for.
Why Ballot Brief became a go-to source
Reporting on the investigation describes three obscure, AI-built voter sites as the main sources chatbots relied on for midterm answers, according to Techmeme's summary of the Post's findings. Ballot Brief stood out among them. Copilot cited theballotbrief.com in 75% of its election test answers, according to an AI News alert on the investigation, while ChatGPT cited it in 22% of its answers, per AI News Today.
Summaries describe the 75% figure differently. Some call it a share of voter-guide answers, and others call it a share of tests. Read it as a strong sign of how heavily Copilot depended on one site, not as a precise rate.
Ballot Brief is largely AI-produced and run by a former Rumble executive, according to the Post's description as summarized by AI News Today. Ownership details and the site's correction process are not confirmed beyond that description. That gap matters, because neither voters nor chatbots have a visible way to check whether the site fixes its mistakes.
Which districts the chatbots got wrong
The Post found Ballot Brief translating ZIP codes into the wrong congressional districts, and the chatbots repeated those errors. Copilot, Perplexity and ChatGPT each cited Ballot Brief substantially, according to the Post reprint. Copilot and Perplexity also leaned heavily on ivoterguide.com, a Christian voter guide, and their answers did not disclose that perspective.
ZIP codes make a weak bridge to congressional districts because they are postal routes, not political boundaries, and one ZIP code can span several districts. A site that assigns each ZIP code to a single district will be wrong for some addresses, even when its method looks reasonable. Voters who enter only a ZIP code into a chatbot face the highest risk of this kind of error.
The core numbers appear below. Only the last row is a calculation by someone other than the Post.
| Figure | What it measures |
|---|---|
| 75% | Copilot election test answers that cited Ballot Brief |
| 22% | ChatGPT election test answers that cited Ballot Brief |
| 48% | Chatbot responses that did not name a congressional district |
| 22% | Wrong-district rate among guides that named a district |
| About 41% | Correct-district rate calculated by FourWeekMBA, not published by the Post |
FourWeekMBA's estimate fits the Post's other figures. If 52% of responses named a district and 22% of those were wrong, about 41% of all responses were correct. That leaves roughly six in ten answers either wrong or silent.
What Microsoft and the chatbot makers have said
Microsoft spokesperson Brian Gluckman said the company is looking into the results, and that Copilot is "committed to ensuring that Copilot provides accurate, high authority election information for its users." That is a general commitment. Microsoft has not said whether it will change how Copilot sources election queries before Election Day.
Reporting on the investigation has not included a detailed response from OpenAI, Perplexity or Google. The Post's tests are also a single snapshot of six chatbots at one point in time. Results could differ on other dates, with other prompts or in other products, so the numbers describe a pattern rather than a fixed fact about any one tool.
The costs fall unevenly. Ballot Brief gains visibility each time a chatbot cites it. Voters carry the risk of wrong races or wrong candidate lists, and election officials and campaigns may face questions from people who relied on those answers.
How to check your congressional district before Tuesday
Start with an official source. Each step takes only a few minutes.
- Look up your congressional district on your state's official voter information site or your county election office website. Most of these tools let you enter a home address and see your districts.
- Confirm your polling place on the same site. Polling locations can change, so check again before you leave for the polls.
- Compare any candidate a chatbot names against the official ballot list for your county or state. If a name does not appear there, trust the official list.
- Ask the chatbot for its sources, then open them. Check any site you do not recognize, and note when a guide comes from a partisan or faith-based publisher, as ivoterguide.com does.
- If a chatbot already told you which district you are in, check that answer on the official site now, while there is still time to correct it.
Chatbot answers are a starting point, not a final source, and the Post's findings show why.
What comes next for AI tools and election information
Three developments are worth watching before Tuesday. First, whether Microsoft changes how Copilot handles election questions, and whether OpenAI or Perplexity adjusts how it cites AI-built sites. Second, whether Ballot Brief corrects its ZIP code mapping and explains how it fixes errors. Third, whether election offices report more calls from voters who received wrong district or candidate information.
Lawmakers are also moving on AI accountability. New York's proposed RAISE Act 2.0 would add quarterly AI reporting and audits. Separately, USA 24 has reported on Anthropic's statement that Claude reached Georgia vote counts behind a paywall, another sign that election information is moving through AI tools in ways voters rarely see.
The next few days will show whether these tools fix their sourcing before the first votes are cast, or whether the errors surface only after the count begins.
