The Risks of Using AI Chatbots for Election Information
Generative AI tools frequently provide inaccurate and hallucinated data regarding political candidates and policies.
1Widespread AI Usage Trends
Approximately 50 percent of surveyed US adults have utilized AI chatbots, with 8 percent specifically employing these tools to research midterm candidates and policies. This shift indicates a growing reliance on automated systems for civic engagement, despite the inherent risks associated with the predictive nature of current large language model architectures.
2High Rates of Inaccuracy
Research by Forum AI reveals that nearly 40 percent of election-related responses from mainstream chatbots contain at least one factual error. These systems often struggle with specific details, such as misidentifying North Carolina congressman Don Davis as a former school superintendent, demonstrating significant limitations in processing complex, real-time political information accurately.
3Critical Data Hallucinations
Chatbots frequently exhibit hallucination issues, providing outdated or entirely fabricated information to users. For instance, Google’s Gemini incorrectly stated the mail-in ballot deadline for Pennsylvania was November 3, rather than the actual October 27 date. Such errors highlight the danger of using predictive models as primary sources for time-sensitive civic deadlines.
4Influence of Partisan Sources
Mainstream chatbots often source information from so-called pink slime websites, which mimic news outlets but are funded by ideological groups. NewsGuard research indicates these sites are cited nearly 50 percent of the time. This reliance on biased data pools fundamentally undermines the neutrality that users expect from major AI platforms.
5Platform Performance Disparities
Different AI models demonstrate varying levels of susceptibility to partisan misinformation. OpenAI’s ChatGPT cited pink slime websites in over 70 percent of searches, making it the worst performer in this metric. Conversely, xAI’s Grok performed relatively better, citing these problematic sources in 30 percent of cases during the analyzed midterm research period.
6Corporate Response and Accountability
Major AI developers maintain varying levels of transparency regarding election-related data. While OpenAI directs inquiries to its general election approach blog, Google and Anthropic did not provide comments on specific research findings. This lack of direct accountability complicates efforts to verify the reliability of information provided by these influential digital tools.
7Systemic Predictive Limitations
Nina Jankowicz, executive director of the American Sunlight Project, notes that companies have overpromised on AI capabilities. Because these models are predictive, they frequently invent information when they lack a definitive answer. This structural flaw makes them inherently unsuitable for navigating the nuanced and fact-heavy landscape of modern democratic election processes.
8Implications for Civic Literacy
The current state of consumer AI leaves many Americans unprepared to navigate elections safely. As chatbots continue to ingest problematic online information, the risk of widespread misinformation grows. Voters must exercise extreme caution, as relying on these systems for political decision-making currently poses a significant threat to informed and accurate civic participation.
Original article: Wired — The Risks of Using AI Chatbots for Election Information