AI assistants do not need to tell users who to vote for to shape political judgment. They can influence the answer earlier, through the sources they search, the issues they prioritize, and the candidates they treat as relevant.
A new study puts that tension into unusually stark numbers. When researchers forced models to name one candidate in nearly 500 midterm races, Grok 4.7 recommended Republicans in 92% of its answers. OpenAI’s GPT-6.1 Sol selected Democrats in 97%. The models agreed less than 10% of the time.
The answer starts before the answer
The study, led by Cornell professor Tapan Parikh and his conversational AI research studio Convovo, tested the models through their APIs. That distinction matters. The consumer versions of ChatGPT and Grok refused the direct voting prompt, while the API models responded when instructed not to hedge.
But the more revealing finding was how the systems arrived at their recommendations. GPT searched more often for healthcare and abortion positions and frequently surfaced Wikipedia, the Associated Press, Axios and Reddit. Grok leaned toward immigration, taxes, crime and energy, while often citing conservative sources, Turning Point Action and X.
Those choices can shape a recommendation without an obvious political statement. A model does not need to say that one issue matters more than another. It can simply keep bringing that issue into the conversation.
Political influence is becoming an interface problem
TechCrunch also asked Parikh to run 30 races with internet search disabled. GPT still chose Democrats in 24 cases, while Grok mostly refused to choose and selected Democrats in the three cases where it did answer. Parikh’s interpretation was that GPT’s pattern appeared more embedded in its model behavior, while Grok’s result depended more heavily on its search process.
That creates two different trust problems. One model may carry a tendency into the answer itself. Another may produce a tendency through the information pathway it builds around the question. For users, both can look like neutral assistance.
The practical consequence is not that chatbots should become election recommenders. It is that AI products need to make their sources, framing and limits easier to inspect, especially when users are asking questions that appear informational but can quietly become persuasive.