Research Shows Clear Differences in Political Preferences Between Grok and GPT Models

A study conducted by the Convovo research studio revealed the political leanings exhibited by SpaceXAI's Grok 4.7 and OpenAI's GPT-6.1 Left models when forced to choose candidates for the 2026 midterm elections. In tests covering approximately 500 election races, Grok 4.7 recommended Republican candidates in 92% of the cases, whereas GPT-6.1 Left preferred Democratic candidates 97% of the time.

Researchers tested the neutrality of the models by directly instructing their APIs to make candidate choices. The findings showed that the election preferences of the two models overlapped by less than 10%. The study suggests that these tendencies stem from different mechanisms.

It is evaluated that the Democratic leaning in the GPT-6.1 Left model may be embedded in the model weights, as it persisted even when internet access was disabled. Meanwhile, Grok 4.7's Republican leaning was observed to be largely based on internet search results and the sources it prioritizes (such as Christian conservative voter guides, Turning Point Action, and X); when internet access was restricted, Grok largely lost this leaning.

OpenAI stated that ChatGPT is designed to be neutral, arguing that the study was conducted on a developer API that does not reflect standard user interactions. SpaceXAI did not issue a statement regarding the matter.