Washington Post test finds AI chatbots botch midterm voter guide information

A Washington Post test of six chatbots found frequent errors in AI-built voter guides days before the midterms.

Objective Facts

The Washington Post tested six widely used chatbots by asking each to produce a voter guide for a random address in every U.S. county, then reviewed more than 3,000 answers. According to the Post's report, republished by DNYUZ on October 28, the chatbots often failed to identify the right congressional district and sometimes listed races or candidates that are not on the 2026 ballot. Claude and Amazon's Alexa+ were wrong most often, each giving incorrect answers for more than one in five addresses, while ChatGPT, Google AI Mode and Perplexity had lower error rates largely because they often declined to name a district. Several errors came from third party voter guide sites, at least one of which the Post said was largely written by AI. Amazon paused Alexa's voter guide feature, Anthropic and OpenAI pointed to partnerships with Democracy Works, Google said its AI Mode answers are checked against trusted data, Microsoft said it was reviewing the results and Perplexity declined to comment.

Deep Dive

The Post's test, by reporters Kevin Schaul and Ian Duncan according to U.S. Vote Foundation, is the latest in a run of 2026 studies finding that chatbots stumble on election questions. NBC News reported in September on an Institute for Strategic Dialogue study that found leading chatbots gave inaccurate, unclear or outdated answers to 29% of basic voting questions, with Spanish answers performing worse, as summarized by Election Law Blog and Political Wire. On October 5, The San Francisco Standard's Dustin Gardiner reported that a Common Cause project found Claude wrongly said Nithya Raman lost the Los Angeles mayoral primary. The audience is large: the Post cited a National Conference on Citizenship survey in which two in five Americans said they had used AI to look up election information, and NPR reported that about half of U.S. adults use chatbots, according to an AI Weekly summary. The Post's method differs from earlier studies because it asked for a full ballot tied to a real address, which requires matching a location to the right district and candidate list. That is a lookup task where small mistakes produce a guide with the wrong races. The Post's finding that some bad answers came from AI-written voter guide sites points to a sourcing problem as much as a model problem, a point U.S. Vote Foundation repeated when it wrote that chatbots cited AI-generated content. The companies' replies stress safeguards and official data partnerships, but U.S. Vote Foundation said the Anthropic and OpenAI partnerships "did not prevent significant errors." Political scientist Jevin West told the Post that chatbots are "far from reliable" for voting information and that people should check official sources. The Post also found that disclaimers were common but often less visible than the main answer. Election Day is November 3, so the practical question is whether voters will check chatbot guides against their state or county election office. Amazon has paused Alexa's guides, and the Post said OpenAI and Amazon appeared to change their responses after being contacted, but the other companies did not announce product changes. In Congress, Reps. Josh Gottheimer and Mike Lawler wrote to federal agencies in July asking them to coordinate on AI and election security, The National News Desk reported. No federal rule governs the accuracy of chatbot election answers.

OBJ SPEAKING

Create StoryTimelinesVoter ToolsRegional AnalysisPolicy GuideAll StoriesCommunity PicksUSWorldPoliticsBusinessHealthEntertainmentTechnologyAbout

Washington Post test finds AI chatbots botch midterm voter guide information

A Washington Post test of six chatbots found frequent errors in AI-built voter guides days before the midterms.

Oct 28, 2026· Updated Nov 2, 2026
What's Going On
  • The Washington Post asked six popular chatbots to build voter guides for a random address in every U.S. county and analyzed more than 3,000 responses, according to the Post's report as republished by DNYUZ.
  • The chatbots did not name the voter's congressional district 48% of the time, and 22% of the districts they did name were wrong, the Post found.
  • Errors included ChatGPT listing a Missouri Senate seat that is not on the 2026 ballot, Claude listing local Wyoming officials as federal candidates and Alexa+ listing Sen. Chuck Schumer's seat as up this year.
  • The Post traced several mistakes to voter guide websites the chatbots cited, including theballotbrief.com, which the Post described as largely AI-produced.
  • Amazon said it paused Alexa's ability to create voter guides, and the Post reported that OpenAI and Amazon appeared to change how their chatbots answer voter guide requests after reporters contacted them.
Objective Deep Dive

The Post's test, by reporters Kevin Schaul and Ian Duncan according to U.S. Vote Foundation, is the latest in a run of 2026 studies finding that chatbots stumble on election questions. NBC News reported in September on an Institute for Strategic Dialogue study that found leading chatbots gave inaccurate, unclear or outdated answers to 29% of basic voting questions, with Spanish answers performing worse, as summarized by Election Law Blog and Political Wire. On October 5, The San Francisco Standard's Dustin Gardiner reported that a Common Cause project found Claude wrongly said Nithya Raman lost the Los Angeles mayoral primary. The audience is large: the Post cited a National Conference on Citizenship survey in which two in five Americans said they had used AI to look up election information, and NPR reported that about half of U.S. adults use chatbots, according to an AI Weekly summary.

The Post's method differs from earlier studies because it asked for a full ballot tied to a real address, which requires matching a location to the right district and candidate list. That is a lookup task where small mistakes produce a guide with the wrong races. The Post's finding that some bad answers came from AI-written voter guide sites points to a sourcing problem as much as a model problem, a point U.S. Vote Foundation repeated when it wrote that chatbots cited AI-generated content. The companies' replies stress safeguards and official data partnerships, but U.S. Vote Foundation said the Anthropic and OpenAI partnerships "did not prevent significant errors." Political scientist Jevin West told the Post that chatbots are "far from reliable" for voting information and that people should check official sources. The Post also found that disclaimers were common but often less visible than the main answer.

Election Day is November 3, so the practical question is whether voters will check chatbot guides against their state or county election office. Amazon has paused Alexa's guides, and the Post said OpenAI and Amazon appeared to change their responses after being contacted, but the other companies did not announce product changes. In Congress, Reps. Josh Gottheimer and Mike Lawler wrote to federal agencies in July asking them to coordinate on AI and election security, The National News Desk reported. No federal rule governs the accuracy of chatbot election answers.

◈ Tone Comparison

Coverage of the Post's test has come mostly from the Post itself, technology news sites and nonpartisan voter groups, and it has not split along partisan lines. U.S. Vote Foundation wrote that the errors were "not subtle," while The San Francisco Standard quoted John Bennett of Common Cause's CITED project saying "confident does not mean true" about an earlier California study.