Audit: AI chatbots gave outdated or wrong answers to about 14% of US voting-logistics questions

As Americans voted in the 3 November 2026 midterm elections, a nonprofit audit reported by NPR found that popular AI chatbots often gave flawed answers to practical voting questions.

Findings

The group Innovating for the Public Good (IFPG) tested major chatbots with election questions in October:

• 13.9% of answers to voting-logistics questions were outdated or inaccurate on a stated fact.
• 5.9% contained errors that could mislead a voter — for example, cause them to miss a deadline, go on the wrong day or skip a required step.
• Across all questions, 5.1% of answers had at least one clear factual error, and 10.9% a clear error or misleading claim.

Differences between chatbots

Accuracy depended more on which chatbot was asked than on the question, the researchers said. Claude and Copilot produced three-quarters of the clear errors. On voting questions, Claude was wrong or outdated in about one answer in three, Copilot in one in four and ChatGPT in one in ten, according to the report.

Company response and caveats

An Anthropic spokesperson told NPR that because IFPG queried Claude through an API, some of its safeguards may not have been active; IFPG says its own comparison of API and web answers found them comparable. The results reflect one organisation’s test sample from October and may change as products are updated. Official state or local election websites remain the authoritative source.