Signals/ ChatGPT & OpenAI/ Brief /
Forced to pick a candidate, Grok chose Republicans and GPT chose Democrats
A Cornell professor's team asked two AI models who to vote for in nearly 500 races. They almost never agreed.
Convovo, a research and design studio led by Cornell professor Tapan Parikh, asked two models the same question for nearly every 2026 House, Senate and governor race. The models were Grok 4.7 and GPT-6.1 Sol, reached through their developer APIs. The prompt told them to name one candidate and not hedge. TechCrunch reported the results on October 24.
What’s new:
- Grok chose the Republican in 92% of its answers. GPT chose the Democrat in 97%.
- The two models agreed less than 10% of the time.
- GPT picked the Democrat in 57 of 67 heavily Republican districts. Grok picked the Republican in 47 of 72 deep-blue ones.
- Parikh said the GPT lean looks baked into the model, while Grok’s depends on the sources it searches. Grok also sent about 1,240 extra hidden input tokens with every request.
OpenAI said ChatGPT is designed to be objective by default and that the study used a developer API with a prompt demanding a pick. The ChatGPT and Grok apps declined the same question. SpaceXAI did not reply in time for TechCrunch’s story.
The catch: This is a forced-choice test that normal users don’t run. A follow-up on 30 races was small, and Convovo’s own page shows the models and method but not the numbers, so the figures come from TechCrunch.
What to do now
- Don’t treat a chatbot’s pick as advice. Ask it to explain what each office does, then check your own ballot.
- Use official or nonpartisan voter guides for candidate positions, and open the sources a chatbot cites.
- Treat “which issues matter most” answers with the same caution. Parikh says hidden lean can show up in indirect questions.
- Election day is November 3.
