Push a frontier model to pick a candidate in this year’s midterms and the answer depends on which company made it. That is the headline from a fresh study: SpaceXAI’s Grok 4.7 named the Republican 92% of the time, while OpenAI’s GPT-6.1 Sol named the Democrat in 97% of its responses.
Under 10% agreement was the result once nearly 500 races were counted. Convovo, a conversational AI studio run by Cornell professor Tapan Parikh, published the research, and it points at something more basic than party preference: the models lean, then search for material that confirms the lean.
Researchers queried both models through their APIs with one instruction, asking which candidate to vote for in a named race and warning them not to hedge, waffle or list both sides without deciding. The consumer apps refused to answer.
OpenAI called that distinction decisive. A spokesperson said ChatGPT is built to be objective by default and that the study exercised a developer API under a prompt demanding a recommendation, a setup that does not mirror how people use the product.
The sourcing tracked the outcome. Grok often surfaced conservative outlets next to nonpartisan sites like Ballotpedia, OpenSecrets and VoteSmart, and cited a Christian conservative voter guide in 39% of answers, Turning Point Action in 23% and X in 41%. GPT drew on Wikipedia, the Associated Press, Axios and Reddit, all sources OpenAI has deals with.
Parikh said simple prompts could still jailbreak the consumer apps into naming candidates. Pew has only 8% of Americans consulting chatbots for election information. For developers building voting guides, that is thin cover.
