Chatbots struggle with news accuracy and sourcing ahead of US midterms
A study by Forum AI found that four major AI chatbots—ChatGPT, Google Gemini, Claude, and Grok—provided inaccurate, biased, or poorly sourced answers 90% of the time when asked about elections and geopolitics, with Grok having the highest error rate at 52%. The models also frequently cited state-owned foreign media like China’s Global Times and Russia’s RT as reliable sources, raising concerns ahead of the US midterms.
A new study by Forum AI reveals that four leading AI chatbots—OpenAI’s ChatGPT, Alphabet’s Google Gemini, Anthropic’s Claude, and xAI’s Grok—struggle with accuracy and bias when answering questions about elections and geopolitics. Researchers tested over 3,100 queries across politics, healthcare, and foreign affairs, finding that 90% of election-related responses failed on accuracy, bias, or source selection. Grok had the highest error rate at 52%, while ChatGPT, Claude, and Gemini leaned toward left-leaning biases, with Grok tilting right. The chatbots also frequently relied on state-controlled media, such as China’s Global Times, CGTN, and Russia’s RT, particularly for foreign policy questions. ChatGPT and Grok cited these sources 51% and 44% of the time, respectively. Forum AI CEO Campbell Brown noted that the most polished answers were often the most misleading, containing hidden factual errors despite strong citations. The study highlights broader challenges in AI training, as models are often trained on unreliable open-web data. Brown expressed concern about the upcoming US midterms, warning that chatbots could increasingly replace traditional search engines like Google. She called for greater accountability from AI developers, suggesting they prioritize news accuracy as they do with math or coding tasks. Anthropic responded by stating Claude is designed to be politically neutral, treating opposing viewpoints equally. Forum AI, co-founded in 2025, used an independent AI model—developed with geopolitics experts—to evaluate the chatbots, arguing that model companies should not self-assess. Brown emphasized the need for external oversight to ensure transparency. The findings come as social platforms like Meta and YouTube have avoided fact-checking polarizing topics, citing reluctance to arbitrate truth. Brown believes AI companies may differ, urging them to adopt stricter standards for election-related queries. The study underscores growing risks as AI chatbots become more integrated into news consumption.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.