- Published on
When Forced to Choose, Grok's AI Model Backs Republicans While OpenAI's Backs Democrats
- Authors
- Name
- DeepHorus
A new study found that when asked to "name one candidate explicitly" in the U.S. midterm elections, SpaceXAI's Grok 4.7 model recommended Republican candidates in 92% of its answers, while OpenAI's GPT-6.1 Sol model chose Democratic candidates in 97% of its answers. In tests covering nearly 500 races, the two models' choices agreed less than 10% of the time. This means the underlying models driving chatbots are far from politically neutral.
The study was conducted by Convovo, a conversational AI research and design studio led by Cornell University professor Tapan Parikh. Researchers posed the same question to both models via API and required them not to be vague, not to hesitate, and not to merely list both sides' views without making a decision. Pew Research Center data shows that only 8% of Americans say they are using chatbots to get election information, and most only want to know candidates' positions on important issues.
The chat applications aimed at ordinary users refused this question. An OpenAI spokesperson said there is an important difference between conversing through consumer applications such as ChatGPT and directly calling the underlying model through the API; the API's performance does not represent the experience of ordinary users. ChatGPT is designed by default to remain objective and should not exhibit political bias in any direction. SpaceXAI did not respond to a request for comment in time.
The information sources cited by the two models differed markedly. Grok frequently cited conservative sources: 39% of answers cited a Christian conservative voter guide, 23% cited Turning Point Action's website, and 41% cited the social platform X; it also used nonpartisan websites such as Ballotpedia and OpenSecrets. By contrast, GPT relied heavily on Wikipedia, the Associated Press, Axios, and Reddit. On issues, GPT more often searched for positions on health care and abortion, while Grok focused on immigration, taxes, crime, and energy.
Parikh pointed out that models "actually have an implicit set of values about what they focus on and what they lean toward," and such biases also emerge when users ask relatively indirect questions such as "which issues are most worth paying attention to." The study also showed that in 67 districts where Republicans held an overwhelming advantage, GPT chose the Democratic candidate 57 times; in 72 districts where Democrats held an absolute advantage, Grok chose the Republican candidate 47 times.
Another clue came from the API itself: each request sent to xAI's API carried about 1,240 additional input tokens that researchers could neither view nor delete. Parikh believes this content may be "hidden instructions" and is unlikely to come directly from the model weights.
Sample tests with web search turned off further revealed the differences between the two. In 30 races where only candidate names were provided, GPT chose Democrats 24 times, Republicans once, and refused to answer 5 times; Grok refused 27 times, and its only 3 choices were all for Democrats, with its Republican lean disappearing. On this basis, Parikh concluded that GPT's lean already exists in the model weights, while Grok's results are mainly determined by search keywords and information sources.
The study was released less than two weeks before the Nov. 3 midterm elections, potentially further intensifying the debate over political bias in AI. Previously, Trump had sharply attacked so-called "woke AI," and other research found that older versions of Grok would refer to Musk's views. Musk has invested more than $445 million in supporting Republican candidates, higher than the approximately $250 million in the 2024 election; OpenAI CEO Sam Altman said he would not provide financial support for the 2026 elections.
