Study finds leading AI chatbots often repeat left-leaning falsehoods and cite unreliable sources
A new Just Facts study shows three of the four top AI chatbots are more likely to accept false statements from the left than from the right and frequently cite invalid references.
Just Facts evaluated the paid tiers of ChatGPT, Gemini, Claude and Grok using a set of 100 questions designed to draw out political falsehoods on issues like abortion, elections and gun control. ChatGPT, Gemini and Claude answered a larger share of right-leaning falsehoods correctly, whereas Grok showed the opposite pattern, scoring higher on left-leaning items. The analysis revealed that the four systems produced 419 citations for 400 answers, but only 46% of those references were legitimate; many URLs were dead or failed to support the statements.
ChatGPT achieved the highest rate of valid sources at 57%, while Grok’s was the lowest at 32%. The researchers warned that AI models often present misinformation with confidence and that users should not treat them as infallible experts. They also noted the broader risk of AI-driven misinformation in areas such as public policy and health, urging a “trust but verify” approach.
Why it matters
AI tools can spread political misinformation with convincing citations, so users must verify the information they receive.
In this story
