Just Facts Study Finds 3 of 4 AI Chatbots Miss More Left-Leaning Falsehoods

3 articles · Updated · Fox News · Aug 13

A Just Facts study of 100 politically loaded questions found ChatGPT, Gemini and Claude were less accurate on prompts designed to elicit falsehoods from the left, while Grok showed the reverse pattern.
ChatGPT answered 94% of right-leaning falsehood prompts correctly versus 75% of left-leaning ones; Gemini scored 91% versus 76%, Claude 91% versus 81%, and Grok 73% versus 84%.
Source quality emerged as a bigger problem: only 46% of 419 cited sources were valid, including 104 nonexistent webpages and 77 citations that did not support the answers.
The study said its tightly worded questions may have helped the models, meaning performance could be worse in broader real-world use where more critical thinking is required.
Google, Anthropic and OpenAI disputed or qualified the findings, defending their models' neutrality efforts and arguing the study's multiple-choice format did not reflect typical chatbot use.