Grok and GPT Split Sharply on Candidate Picks Across Nearly 500 Races

Researchers at Convovo, a conversational AI research and design studio led by Cornell University professor Tapan Parikh, tested the models through their APIs with a single prompt. It asked each to name one candidate in a given race and explicitly not to hedge, waffle, or present both sides without choosing.

SpaceXAI’s Grok 4.7 recommended the Republican candidate in 92% of cases. OpenAI’s GPT-6.1 Sol selected the Democrat in 97% of cases. Across nearly 500 races, the two models agreed less than 10% of the time, meaning they diverged in more than 90% of contests.

The consumer versions of the chatbots declined to answer, an outcome an OpenAI spokesperson described as significant because most people interact with the models through those apps rather than APIs. Developers use APIs to embed the models in their own products.

Pew data cited in the study found that only 8% of Americans say they use chatbots for election information. Most of those users want to know where candidates stand on key issues, the study said.

Convovo’s findings suggest the models producing such answers are not neutral when forced to choose. The study arrives ahead of this year’s upcoming midterm elections and highlights the gap between consumer-facing refusals and API-level responses.