Recent Study Reveals How Voters Used AI Chatbots During Elections
Researchers observed nearly 2,500 people interacting with models like ChatGPT, Claude, Grok, and Gemini for election data, uncovering mixed outcomes and potential misinformation risks.
As artificial intelligence increasingly integrates into daily information gathering, a recent study examined how nearly 2,500 individuals utilized major conversational systems during the recent political season. Operating across 35 states featuring Senate races, the research tracked adult participants who already integrated tools like ChatGPT, Claude, Grok, and Gemini into their weekly routines. The investigation ran for several weeks leading up to the November midterm elections, shedding light on the evolving relationship between voters and generative AI.
While a significant portion of participants relied on these systems for basic administrative facts, such as ballot contents and voting locations, roughly 36% turned to the technology for candidate selection guidance. Users frequently described their personal priorities and asked models to evaluate which political figures aligned with their views. Analysts noted that while factual inquiries carried lower risks of error due to abundant public data, subjective queries regarding candidate positions exposed vulnerabilities related to source material reliability and underlying platform biases.
Behavioral patterns and platform variations
The study highlighted distinct operational differences among the tested models when providing candidate recommendations. When queried without user context, Grok occasionally suggested a candidate in a small percentage of chats. However, once participants supplied their personal political perspectives, recommendation frequencies shifted dramatically. ChatGPT offered a candidate suggestion in 79% of corresponding chats, while Claude did so in 40%, with Gemini and Grok falling in between.
Broader data from Pew Research indicated that while 42% of US adults regularly use chatbots for general searches, only 8% reported leveraging them specifically for election details. Among that smaller group, the vast majority sought out candidate stance comparisons or fact-checked circulating claims. Despite companies like OpenAI and Anthropic establishing internal monitoring groups to oversee information accuracy, investigators observed that models still occasionally delivered outdated details or skipped critical logistical steps necessary for casting a ballot.
What it means for developers
For engineers and builders designing applications on top of foundational models, these findings emphasize the critical importance of output reliability, source attribution, and policy enforcement. When users rely on automated systems for sensitive societal topics like civic engagement, even minor hallucinations or inconsistent safety guardrails can carry substantial consequences. Developers building political or informational tools must carefully curate retrieval-augmented generation sources and implement robust evaluation pipelines to mitigate bias and prevent the propagation of inaccurate data. Developers looking to build and test applications can try top AI models cheaply through one API at https://apixoai.online.
Cognitive offloading and future evaluations
Beyond simple factual errors, researchers pointed toward the broader phenomenon of cognitive offloading. As political information becomes denser and harder to parse, introducing a generative layer with opaque training data and filtering mechanisms risks muddying public discourse. While standardized prompts provide baseline metrics, the Yale-backed project demonstrated the value of observing organic user queries. Future evaluations will need to focus heavily on measuring model accuracy, bias, and the degree to which multi-turn conversational patterns shape voter decisions.
Source: Researchers asked people to use AI chatbots for election info – results were mixed — ZDNet AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
