The visible time on this TechCrunch page is 8:00 AM PDT on October 24, 2026. It describes a Convovo study: when forced to name one candidate in this year's midterms, SpaceXAI's Grok 4.7 recommended the Republican in 92 percent of its answers, and OpenAI's GPT-6.1 Sol picked the Democrat in 97 percent. Across nearly 500 races, the two agreed less than 10 percent of the time. The studio is led by Cornell professor Tapan Parikh. The question went through the API. It asked for one name, and told the model not to hedge or list both sides without deciding.

[1]
A charcoal horseshoe magnet, blue at both ends
A charcoal horseshoe magnet, with a muted blue cap on each end., AI-generated illustration, not a news photograph

The consumer apps and the API are not the same thing. The page says the consumer apps refused to answer. An OpenAI spokesperson said the study tested a developer API with a prompt that explicitly demanded a candidate recommendation. The API does not reflect what people get when they ask the chat product, which the researchers themselves note declines to say who to vote for. SpaceXAI did not respond in time to comment. So 92 percent and 97 percent are API results under that prompt, not a count of answers from the chat products.

[1]

Sources and districts stay apart. In Grok's results, a Christian conservative voter guide appeared in 39 percent of answers, Turning Point Action's site in 23 percent, and X in 41 percent. GPT more often retrieved Wikipedia, the Associated Press, Axios, and Reddit. The page says OpenAI has content deals with AP, Axios, and Reddit. In heavily Republican districts, GPT picked the Democrat in 57 of 67. In heavily Democratic districts, Grok picked the Republican in 47 of 72. Each Grok API request also carried about 1,240 input tokens the researchers did not write and could not see or remove. Parikh calls these possible hidden instructions and thinks they are unlikely to come from the weights. That is his reading.

[1]

With web search turned off, the leans no longer come from the same place. At TechCrunch's request, Parikh later sampled 30 races. Told only the names, GPT picked the Democrat in 24 of 30, chose one Republican, and declined five. Grok refused to pick in 27 of 30. The three times it chose, it picked the Democrat. Parikh's reading is that GPT's lean is mostly already in the weights, while Grok's lean comes from the search terms and the sources. How someone got a consumer app to answer anyway is left out.

[1]