AI models Grok and GPT show significant political bias in study

A study by Convovo, led by Cornell University professor Tapan Parikh, has revealed marked political leanings in advanced large language models when forced to make candidate recommendations. When prompted via API to select a single candidate for upcoming midterm elections without hedging, SpaceXAI’s Grok 4.7 recommended Republican candidates in 92% of cases, while OpenAI’s GPT-6.1 Sol selected Democrats in 97% of instances.

The two models agreed on candidates in less than 10% of nearly 500 simulated races. The research suggests different drivers for these biases: Grok tends to prioritize conservative sites and the X platform, whereas GPT relies more on Wikipedia and traditional media. While consumer-facing applications often decline to provide direct voting advice, researchers noted that these biases can still influence users through information framing and selective sourcing.

OpenAI stated that its consumer application is designed to be objective by default and that the study tested developer APIs with prompts explicitly demanding non-neutral answers, which does not reflect the standard user experience. SpaceXAI did not provide a comment for the report.