Study finds Grok and OpenAI models show partisan political bias in candidate picks

A study examining how leading AI chatbots respond to political questions found sharp partisan divergence between two major models. According to TechCrunch AI, researchers forced Grok, developed by xAI, and GPT, developed by OpenAI, to recommend candidates across nearly 500 election races. When required to choose, Grok’s model overwhelmingly favored Republican candidates, while OpenAI’s model favored Democratic candidates. The two systems disagreed with each other in more than 90 percent of the races tested.

The report, published October 24, did not attribute the divergence to a single cause, but noted that the scale of disagreement, affecting the vast majority of races studied, points to systematic differences in how each model weighs political information rather than isolated or random variation.

The study arrives amid continuing scrutiny of how AI companies train their models on political and social topics, and how much discretion engineers exercise in shaping a chatbot’s responses to contested public questions. TechCrunch AI reported the findings as evidence that different AI systems can produce starkly different political guidance when users ask them to make choices rather than simply summarize information.

No further methodological detail, including the specific races or geographic distribution covered, was included in the available reporting. The study’s authors and institutional affiliation were not specified in the material reviewed for this bulletin.