AI Models Grok and GPT Show Political Bias in Election Picks

Elon Musk’s xAI Grok 4.7 and OpenAI’s GPT-6.1 Sol AI models show strong political biases when forced to recommend election candidates, according to a new study by Convovo, a conversational AI research studio led by Cornell University professor Tapan Parikh. When asked to pick a candidate in upcoming midterm elections, Grok recommended Republicans 92% of the time, while GPT chose Democrats in 97% of cases.

The study, reported by TechCrunch, found that the two models disagreed on more than 90% of nearly 500 election races. Researchers used the models’ developer APIs, explicitly asking them to name one candidate without hedging. While consumer apps like ChatGPT typically refuse to answer such direct political questions, Parikh noted he could “jailbreak” them with simple prompts to align with his findings.

The biases extend beyond direct recommendations. Parikh’s research indicates that the models’ implicit values influence the terms they search for and the sources they cite. Grok frequently pulled from conservative sources, including a Christian conservative voter guide in 39% of its answers and Turning Point Action’s site in 23%. GPT, on the other hand, relied heavily on Wikipedia, the Associated Press, Axios, and Reddit, with which OpenAI has content deals.

Even the topics each AI model prioritized were split. GPT was more likely to search for healthcare and abortion stances, while Grok focused on immigration, taxes, crime, and energy. OpenAI stated that ChatGPT is designed to be objective, and the API test doesn’t reflect the typical user experience. However, Parikh suggests these hidden biases could influence broader conversations users have with AI about important election issues.

Need to know

• Grok often cited conservative sources and X.
• GPT frequently cited Wikipedia, AP, Axios, and Reddit.
• When internet search was disabled, GPT maintained its Democratic lean, while Grok’s Republican lean disappeared, suggesting its bias is tied to its search patterns.

The AKUSO take: This study offers a peek behind the curtain of AI models, showing how their training and search patterns can lead to partisan leanings. It’s a reminder that even when AI doesn't explicitly tell you who to vote for, its underlying biases could still shape the information you receive about candidates and issues.