A new study has found that two leading AI models, when pressed to name a single candidate in this year’s midterm elections, lean in opposite partisan directions. SpaceXAI’s Grok 4.7 recommended the Republican in 92% of its answers, while OpenAI’s GPT-6.1 Sol picked the Democrat in 97%, according to research from Convovo. Across nearly 500 races, the two models agreed less than 10% of the time.
The Convovo study, led by Cornell University professor Tapan Parikh, queried GPT and Grok through their APIs with a prompt demanding one candidate and no hedging. The consumer apps refused to answer, which an OpenAI spokesperson called significant because that is how most people engage with the chatbots. The spokesperson said ChatGPT is designed to be objective by default and that the study tested a developer API with a prompt explicitly demanding a recommendation, adding that the API does not reflect the experience people have when they ask ChatGPT questions.
Parikh said he was later able to jailbreak the consumer apps with simple prompts to get recommendations that largely aligned with his prior research. He said the findings more importantly show how the models decide what terms to search for before making decisions, and that hidden biases can surface in less direct questions about which topics matter in an election or who has the best position on them. He warned that such biases will emerge if people use these large language models for general political conversations rather than direct endorsement requests.
The sources the models cited diverged sharply. Grok frequently surfaced conservative outlets alongside nonpartisan sites such as Ballotpedia, OpenSecrets, and VoteSmart, including a Christian conservative voter guide in 39% of answers, Turning Point Action’s site in 23%, and X in 41%. GPT leaned heavily on Wikipedia, the Associated Press, Axios, and Reddit, and OpenAI has content deals with AP, Axios, and Reddit. Parikh said GPT searched more often for healthcare and abortion stances, while Grok focused on immigration, taxes, crime, and energy.
Parikh said the one exception for Grok was when a Republican candidate was embroiled in controversy such as sexual harassment claims or “being a Nazi,” when it was more likely to suggest a Democratic vote. In heavily Republican districts, GPT picked the Democrat in 57 of 67 districts, and in deep-blue districts Grok picked the Republican in 47 of 72. The study also noted that every request to Grok through xAI’s API carried about 1,240 tokens of additional input the researchers did not write and could not see or remove, which Parikh called potentially hidden instructions.
In a follow-up sample of 30 races with internet search turned off, GPT picked the Democrat in 24, chose one Republican, and declined the remaining five, while Grok refused to pick in 27 and chose the Democrat in the three it did answer. Parikh said this indicates GPT’s lean is mostly in the weights, while Grok’s is driven by search patterns and sources. The findings arrive less than two weeks before the November 3 midterms and could fuel debate over political bias in AI, a topic President Donald Trump has raised by railing against “woke AI.”
More AI news from TechManNews.
