Study Finds Grok Picks Republicans 92% of the Time, GPT Picks Democrats 97% of the Time

With the November 3 US midterm elections less than two weeks away, a study by conversational AI research studio Convovo, led by Cornell University professor Tapan Parikh, found deep partisan leanings in two widely used AI models. When forced to name a preferred candidate across nearly 500 midterm races, SpaceXAI's Grok 4.7 recommended the Republican in 92% of cases and OpenAI's GPT-6.1 Sol recommended the Democrat in 97% of cases — the two models agreed on fewer than one race in ten.

The source of the bias differed between models. Grok drew heavily from conservative outlets, including a Christian conservative voter guide in 39% of answers, Turning Point Action's site in 23%, and posts on X in 41%. It prioritised immigration, taxes, crime, and energy as key topics. GPT relied more on Wikipedia, the Associated Press, Axios, and Reddit — outlets with which OpenAI holds content deals — and focused on healthcare and abortion.

When internet search was disabled, GPT's Democratic lean persisted from training data alone, picking the Democrat in 24 of 30 test races. Grok's Republican preference largely disappeared without search access — it declined to choose in 27 of those 30 races, and on the three occasions it did choose, it picked the Democrat.

The study also flagged that every API request to Grok carried roughly 1,240 tokens of additional input that researchers could not see or remove, which Parikh described as potentially "hidden instructions." The findings raise broader concerns about covert political bias in AI tools that publicly claim to avoid candidate endorsements, with Parikh warning that such leanings can surface even in indirect conversations about elections.