SpaceXAI’s Grok 4.7 backed the Republican in 92% of its answers and OpenAI’s GPT-6.1 Sol chose the Democrat in 97% when told to pick one name in nearly 500 midterm races, according to a study from Convovo, a research studio led by Cornell professor Tapan Parikh. The researchers queried both models through their APIs, since the ChatGPT and Grok apps decline to make picks, which OpenAI said makes the results unrepresentative of how people use ChatGPT. When Parikh reran 30 races with web search off, GPT still chose the Democrat 24 times while Grok declined to choose in 27, which he said means GPT’s lean is built into the model and Grok’s comes from what it searches. With search on, a voter guide aimed at Christian conservatives turned up in 39% of Grok’s answers, and X in 41%.
