Voters increasingly rely on artificial intelligence chatbots to gather election information, but studies indicate these systems frequently deliver misleading or outdated guidance that could affect voting participation, according to recent research. A Pew Research Center survey conducted earlier this year found that approximately half of American adults use AI-powered chatbots, with 42% employing them to search for information. Innovating for the Public Good, a nonprofit focused on democracy research, tested seven major chatbot platforms—including Google AI Mode and AI Overviews, Anthropic’s Claude, Microsoft Copilot, Google Gemini, OpenAI’s ChatGPT, and Perplexity—on their ability to answer authentic voter questions.
Data collected in October revealed substantial accuracy problems. The research found that 13.9% of chatbot answers to voting-logistics questions contained outdated or inaccurate information, while 5.9% of responses included errors that could mislead voters and potentially cause them to miss registration deadlines or vote on incorrect dates. Across all question categories, 5.1% of answers contained clear factual errors and 10.9% included either clear errors or misleading claims. Notably, accuracy varied significantly by platform, with Claude and Copilot producing three-quarters of identified errors. Claude provided incorrect information one in three times on voting questions, Copilot one in four times, and ChatGPT one in ten times.
The platforms have acknowledged these concerns and indicated they are implementing improvements. Microsoft stated it has made enhancements to how Copilot responds to election-related questions, while Anthropic, Meta, OpenAI, and Google indicated they are working to prevent election disinformation. Anthropic suggested that some of Claude’s safeguards may not have been fully active during the research, which was conducted using an Application Programming Interface rather than the standard web interface.
Researchers also identified emerging tactics to exploit chatbots. Political campaigns and influence operations are now creating content specifically designed to be picked up by chatbots through a process called generative engine optimization. Additionally, chatbots frequently cited prediction market websites rather than traditional polling data when answering questions about race outcomes, potentially presenting betting odds as equivalent to legitimate poll results.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI