Most AI chatbots won't tell you who to vote for, but a new study found that when pushed, the models behind them lean in opposite directions. In a test of nearly 500 races ahead of the November 3 midterms, xAI's Grok 4.7 recommended the Republican candidate in 92% of answers, while OpenAI's GPT-6.1 Sol picked the Democrat in 97%, per TechCrunch. The two agreed less than 10% of the time.

The research comes from Convovo, a conversational AI studio led by Cornell professor Tapan Parikh. Researchers queried the models through their APIs with a prompt demanding one candidate, no hedging. The consumer apps refused to answer—a point OpenAI stressed, saying an API test with a forced-choice prompt doesn't reflect how people use ChatGPT, which 'is designed to be objective by default.' xAI did not respond to a request for comment.

The study also examined sourcing and search behavior. Grok surfaced conservative outlets alongside nonpartisan sites, and GPT leaned on Wikipedia, the AP, Axios, and Reddit. When search was disabled in a smaller 30-race sample, GPT still favored Democrats, while Grok mostly declined to choose. Parikh said that suggests GPT's lean is baked into its training, while Grok's depends on what it searches for.

The findings arrive amid growing scrutiny of political bias in AI, a debate President Trump has fueled with attacks on 'woke AI.' The study's methodology—forced-choice API prompts rather than organic use—limits how far the results can be generalized.

Source: TechCrunch