The Washington Post asked six popular chatbots to build midterm voter guides for a random address in every US county, and the guides often got the basics wrong, according to the Post’s analysis of more than 3,000 of them published on 28 October.
The chatbots failed to say which congressional district an address was in 48 percent of the time, and when they did name one they were wrong 22 percent of the time, the Post found. That means about 41% of guides, roughly 4 in 10, named the voter’s correct district, by our arithmetic.
Short explainer · Chatbot voter guides: right district 41%
A vertical news Short summarising this story. Findings come from The Washington Post’s 28 October test; the 41% is our arithmetic.
Business Pill · The Model and the Harness
A short explainer on how the software around a model shapes what it does. It teaches the general idea only and says nothing about any company in this piece.
The key insight: As we read the Post’s test, the weak link was not the model but what it read: small, partly AI-written voter sites outranked official tools inside the answers, and only about 41% of guides named the right congressional district, by our arithmetic.
The Post says ChatGPT included a US Senate seat that is not on the ballot this year for an address in Missouri, and Anthropic’s Claude listed local politicians as running for federal office for an address in Wyoming. Google’s AI Mode placed a Maryland address in the wrong district and said Rep. Steny Hoyer faced a challenger; the Post notes Hoyer is not running for reelection.
Claude and Alexa+ were wrong most often, the Post reports, “both responding incorrectly for more than 1 in 5 addresses”. ChatGPT, Google AI Mode and Perplexity were wrong the least often, “but only because they typically did not respond with a district at all”.
How six chatbots’ midterm voter guides handled the congressional district, from The Washington Post’s test of a random address in every US county (requests sent 23 September 2026). The 48% and 22% are the Post’s; the split into 52%, 41% and 11% is our arithmetic.
Where the Errors Came From
The Post traced some errors to voter-guide websites, many of which appear to be AI-generated. ChatGPT cited myvotega.com for an address in Burke County, Georgia; the site listed the county in the wrong district. Its operator told the Post: “I’m just a guy with a laptop”, and has since fixed the error.
One of the top sites the chatbots drew on was theballotbrief.com, which the Post describes as largely produced by AI tools and run by a former Rumble executive. The Post found it translating Zip codes to the wrong congressional districts, errors then passed on by chatbots. “Copilot, Perplexity and ChatGPT each cited the Ballot Brief substantially”, the Post says. Its founder, Ben Torres Ezrick, told the Post the site draws on official government sources.
Copilot and Perplexity also relied heavily on ivoterguide.com, a Christian voter guide, the Post reports, and the chatbot answers did not carry that context.
What the Companies Said
Amazon said it identified the cause of the errors and has paused Alexa’s ability to create voter guides. A Microsoft spokesperson said the company is “committed to ensuring that Copilot provides accurate, high authority election information for its users”. Anthropic and OpenAI both pointed to partnerships with Democracy Works; Google said its systems corroborate generated claims against trusted data sources. SpaceXAI did not respond and Perplexity declined to comment, the Post says.
The Post notes that Amazon Executive Chairman Jeff Bezos owns the Post, and that the Post has partnerships with OpenAI and Perplexity.
The Business Lever: Answer Engines Inherit Their Sources’ Errors
As we read the test, the chatbots’ weak point was source selection: small, partly AI-written sites outranked official tools inside the answers, and disclaimers sat below long responses. As we read it, that puts the choice of sources at the centre of answer engines’ accuracy on midterm questions. Our earlier coverage of the shared AI-training opt-out and Google’s Connected Apps rules tracks how the same companies decide what their assistants read.
The Structural Read
As we read the results, the chatbots that were wrong least often were the ones that declined to name a district; the Post says ChatGPT, Google AI Mode and Perplexity ranked best only because they typically did not respond with a district at all.
As we read the sourcing section, errors travelled from site to answer: the Post found the Ballot Brief translating Zip codes to the wrong districts, and says Copilot, Perplexity and ChatGPT each cited it substantially.
As we read the company responses, the fixes named are partnerships and pauses: Anthropic and OpenAI point to Democracy Works, and Amazon says it paused Alexa’s ability to create voter guides.
The Washington Post, 28 October 2026
“Copilot, Perplexity and ChatGPT each cited the Ballot Brief substantially”
Three Implications
FOR VOTERS The Post’s test found guides named the correct district only about 41% of the time, by our arithmetic, and disclaimers often sat below long answers.
FOR CHATBOT MAKERS The Post reports OpenAI and Amazon appeared to change how their chatbots answer voter-guide requests after being contacted, and Microsoft says it is looking into the results.
FOR SMALL PUBLISHERS The Post traced errors to small voter-guide sites, one run by a self-described “guy with a laptop”, that chatbots cited directly.
The analysis puts it in one line: “The surface, not the model, governs autonomy.”
As we read it, the Post’s test shows the same split on the answer side: which sources the surface pulls in decided whether a voter guide was right. The framework does not discuss voter guides.
The Post’s requests were sent on 23 September through the analytics firm Promptwatch, and OpenAI and Amazon appeared to change how their chatbots answer voter-guide requests after the Post contacted them, so current answers may differ. The Post excluded Missouri addresses from its district analysis because boundaries were unresolved. The test measured districts, not every race on the ballot.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
In the Post’s test of six chatbots, voter guides named the correct congressional district about 41% of the time by our arithmetic, and some errors traced back to small AI-built voter sites the chatbots cited.
A note on sourcing. Findings, examples and company responses come from The Washington Post’s 28 October 2026 analysis, read in full via syndication; we did not run the test ourselves. The 41% figure is our arithmetic from the Post’s 48% and 22%. We did not contact the companies.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
