A recent study conducted by Convovo has revealed stark political leanings in AI models when asked to pick candidates in the 2026 midterm elections. SpaceXAI’s Grok 4.7 overwhelmingly recommended Republican candidates, choosing them in 92% of cases, while OpenAI’s GPT-6.1 Sol preferred Democrats, selecting them in 97% of races. The two models disagreed on more than 90% of the nearly 500 races studied. Interestingly, these results come from AI model APIs rather than consumer-facing chatbot interfaces, which typically refuse direct endorsements, reflecting a key difference in how the models operate in various contexts.

The discrepancy appears tied to the AI systems’ methods for sourcing and interpreting information. Grok tended to draw from conservative sources like Christian conservative voter guides and Turning Point Action, often searching topics associated with Republican priorities such as immigration, taxes, and energy. GPT’s sources leaned toward mainstream media and public knowledge bases like Wikipedia, Associated Press, and Reddit, with an emphasis on healthcare and abortion issues. Notably, when internet search functionalities were turned off, Grok's Republican bias largely disappeared, suggesting its political alignment is significantly influenced by search behavior and source selection, whereas GPT’s Democratic preference seemed embedded in its training data.

Underlying the study’s findings is the concern that AI models may carry implicit biases or political leanings that influence the information they present, even when operating under the guise of neutrality. Cornell professor Tapan Parikh, who led the study, described these biases as "implicit sets of values" shaped by search patterns and sources. More broadly, previous research warns that large language models can covertly sway users through selective framing and emphasis, raising questions about the transparency and fairness of AI-driven information, especially in politically sensitive contexts.

The study surfaces as the 2026 midterm elections approach, highlighting the complexities around AI’s role in political information dissemination. Companies like SpaceXAI and OpenAI differ in both model design and corporate leadership donations—Elon Musk backing GOP candidates with over $445 million this cycle, and OpenAI’s CEO Sam Altman historically supporting Democrats but sitting out the 2026 races. With AI becoming an increasingly influential source for voters, these findings fuel ongoing debates over political bias in tech and the need for clearer guidelines on AI impartiality and the responsible use of conversational agents during elections.