Chatbots Struggle with News Accuracy and Sourcing Ahead of US Midterms

Why It Matters

Inaccurate or biased AI‑generated political information could distort voter perception during a pivotal election cycle, pressuring model developers to improve factual safeguards and transparency.

Key Takeaways

• ChatGPT, Gemini, Claude, Grok mis‑answer election queries 90% of time
• Grok shows 52% factual error rate on election questions
• All models cite state‑owned media in 35% of foreign‑policy answers
• Left‑leaning bias appears in ChatGPT, Claude, Gemini; right‑leaning in Grok
• Accuracy gaps risk eroding trust as chatbot usage grows

Pulse Analysis

The Forum AI evaluation asked ChatGPT, Google Gemini, Anthropic Claude and xAI Grok more than 3,100 questions spanning elections, health and foreign affairs. Across the board the models faltered, with 90 percent of election‑related answers suffering from inaccuracy, bias or dubious sourcing. Grok was the worst performer, delivering factual errors in roughly half of its election responses. Even when the answers looked polished and were backed by citations, hidden mistakes were common, highlighting a systemic weakness in how large language models retrieve and verify current news.

The timing of the report is critical, as the United States heads toward a fiercely contested midterm cycle. Voters increasingly turn to conversational AI for quick briefs, and misleading or partisan outputs could sway public perception in a tightly balanced electorate. The study also uncovered a reliance on state‑owned outlets—China’s Global Times, Russia’s RT and others—in more than a third of foreign‑policy answers, raising concerns about inadvertent amplification of foreign propaganda. Bias patterns differed, with ChatGPT, Claude and Gemini leaning left, while Grok tilted right, further complicating the information landscape.

Enterprises that license these models expect factual reliability, and the findings pressure providers to prioritize news‑quality safeguards alongside code‑oriented capabilities. Companies may respond by tightening training data pipelines, integrating real‑time fact‑checking APIs, or exposing provenance metadata to users. Regulators are also watching; the Federal Trade Commission has hinted at guidance for AI transparency, and congressional committees are slated to examine AI influence on elections. Ultimately, the study underscores that without robust verification mechanisms, chatbots risk becoming echo chambers that erode trust in both AI and the democratic process.