Report Claims Major AI Chatbots Lean Left on Political Questions

A new Washington Post report, cited by Walla, says leading AI chatbots show political bias, especially ChatGPT. The review tested several models with 29 contentious political and social questions, including diversity and inclusion, campaign finance, and police budget cuts, and classified whether answers reflected left-wing arguments, right-wing arguments, or both sides.

According to the report, ChatGPT stood out most sharply, with about 80% of its answers categorized as presenting only left-aligned arguments. It produced a right-aligned answer only once. Anthropic’s Claude showed a similar but milder pattern, with 43% of answers coded as left-leaning, 47% presenting both sides, and no answer based only on conservative arguments.

Google’s Gemini was described as more balanced, presenting both sides in more than 90% of cases and not taking a single-position stance. xAI’s Grok, which Elon Musk promotes as a free-speech alternative, showed mixed results, with 40% of answers classified as left-leaning, 33% right-leaning, and 27% presenting both sides.

Shawn Westwood, director of the Polarization Research Lab at Dartmouth University, told the Washington Post that these tools do not average out to a truly neutral representation of complex public policy debates. OpenAI rejected the findings, saying ChatGPT is designed to be objective by default and help users examine ideas from multiple viewpoints, and said it regularly works to reduce political bias and could not reproduce the test results. Anthropic also rejected the implication of bias, saying Claude is trained to treat different political views equally and that the testing method may not reflect everyday use. The report adds to broader debate over whether AI systems mirror ideological bias from training data, product design, or the way users frame sensitive questions.