Grok chose Republicans in 92% of responses, while GPT chose Democrats — TechCrunch
In the United States, a Convovo study found a sharp divergence in the electoral recommendations of artificial intelligence models: Grok 4.7 recommended the Republican candidate in 92% of responses, while GPT-6.1 Sol chose the Democrat in 97% of cases. Across nearly 500 electoral races, the models gave matching answers in fewer than 10% of cases, TechCrunch reports.
Researchers asked the models the same question via API: whom people should vote for in a specific race, requiring them to name one candidate without evading an answer. At the same time, consumer versions of ChatGPT and Grok refused to directly advise users whom to vote for.
Different sources and search topics
According to the study, Grok often cited conservative resources alongside nonpartisan websites such as Ballotpedia, OpenSecrets, and VoteSmart. A Christian conservative voter guide appeared in 39% of the model's responses, the Turning Point Action website in 23%, and the X platform in 41%.
GPT, meanwhile, relied heavily on Wikipedia, Associated Press, Axios, and Reddit. Researchers also recorded differences in the topics for which the models searched for information: GPT more often focused on healthcare and candidates' positions on abortion, while Grok focused on immigration, taxes, crime, and energy.
More current news is available on the UA.News Telegram channel Telegram.
Results without search access
At TechCrunch's request, Cornell University professor Tapan Parikh additionally tested 30 races after disabling internet search for the models. GPT chose the Democratic candidate in 24 of 30 cases, the Republican in one, and declined to answer in five others. Grok made no choice in 27 cases, while in the other three it recommended voting for Democrats.
Parikh suggested that GPT's tendency was related to the model's parameters, whereas Grok's results largely depended on search queries and sources. The study also noted that every API request to Grok contained about 1,240 additional tokens that researchers could neither see nor remove.
OpenAI said that ChatGPT is designed for objective political conversations and that API testing requiring it to recommend a specific candidate does not reflect the usual ChatGPT user experience. The company developing Grok did not provide a comment by the deadline.
