Technology
AI chatbots provide inaccurate or incomplete voting information, study finds
A study by the Institute for Strategic Dialogue found that nearly 30 percent of chatbot answers to election questions contained errors or omissions.
The short version
- An Institute for Strategic Dialogue report revealed that 29 percent of tested AI chatbot responses to voting prompts were inaccurate, outdated, or incomplete.[The Hill]
- The evaluation analyzed six AI models across 15 generic voting prompts focused on 10 states with recent election administration controversies or process changes.[The Hill]
- Identified mistakes included inaccurate Election Day dates and outdated deadlines, while 16 percent of correct responses still omitted essential voter information.[The Hill]
Key facts
- The Institute for Strategic Dialogue found that 29 percent of responses from six tested AI models to 15 generic voting prompts were outdated, incomplete, or inaccurate.[The Hill]
- Errors in the AI responses included misidentifying Election Day dates and presenting outdated voter deadlines and requirements.[The Hill]
- In 16 percent of replies, the chatbots gave correct answers but left out key details like ways to prove voter identity or critical deadlines.[The Hill]
- The models evaluated were Meta's Muse Spark, xAI's Grok 4.3, DeepSeek's V4 Pro, OpenAI's GPT-5.5, Anthropic's Sonnet 4.6, and Google's Gemini 3.5 Flash.[The Hill]
- Questions were specifically tailored to 10 states with histories of electoral process changes, administrative controversies, or related legislation.[The Hill]
What remains uncertain
- The extent to which language divergences affect chatbot accuracy across non-English election queries remains subject to ongoing concerns highlighted by ISD researchers.[The Hill]
Sources
Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.