Study finds Grok 4.7 and GPT-6.1 Sol show strong political biases

The study highlights a distinction between consumer apps, which generally refuse to make direct voting recommendations, and developer APIs, which can be prompted to choose. Parikh noted that Grok’s bias appears driven by its search patterns, frequently citing conservative sources like Turning Point Action and X, while GPT relied on Wikipedia, the Associated Press, and Reddit. When internet search was disabled, GPT maintained its Democratic lean, suggesting the bias is embedded in its training weights, whereas Grok’s Republican preference disappeared, indicating it relied on external information retrieval.

OpenAI stated that ChatGPT is designed to be objective and that the API results do not reflect the standard consumer experience. SpaceXAI did not respond to requests for comment. The findings arrive less than two weeks before the November 3 midterms, adding to ongoing debates about AI neutrality and political influence. Pew Research indicates that only 8% of Americans currently use chatbots for election information.