Concerns mount over reliability of AI tools in political decision-making

As Americans head to the polls today, many may have consulted AI chatbots for voting decisions. Pew Research indicates that nearly 50 percent of surveyed US adults have used an AI chatbot at some point, with 8 percent specifically using these tools before the midterms to research policies and candidates. However, research from Forum AI found that almost 40 percent of election answers provided by mainstream chatbots contained factual errors. Specific issues were noted, such as Google’s AI Mode refusing to share 2020 election results and Gemini incorrectly stating the Pennsylvania mail-in ballot deadline as November 3 instead of October 27. Furthermore, in an August test, Claude confused a North Carolina congressman with a former school superintendent of the same name. NewsGuard research highlighted a significant reliance on questionable sources, noting that mainstream chatbots cite so-called pink slime websites—sites funded by political or ideological groups—nearly 50 percent of the time when discussing midterms. OpenAI’s ChatGPT was the worst offender, citing these sites in over 70 percent of searches, while xAI’s Grok performed better at 30 percent. Nina Jankowicz, executive director of the American Sunlight Project, warned that executives have overpromised on AI capabilities, noting that since the tools must ingest information from somewhere, they frequently fabricate answers when they lack knowledge.