Grok votes red, GPT votes blue — forced to pick, they agree on almost nothing

Convovo, a conversational AI research studio led by Cornell professor Tapan Parikh, forced Grok 4.7 and GPT-6.1 Sol to name a candidate in nearly 500 midterm races. Grok picked the Republican in 92% of its answers; GPT-6.1 Sol picked the Democrat in 97%. The two models agreed in fewer than 10% of races — and the lean held even in districts that vote overwhelmingly the other way.

The catch is where the answers came from: the consumer apps refused to pick, so the results are from the developer APIs — the same models companies build products on. OpenAI says the API test does not reflect how people experience ChatGPT, but Parikh says simple prompts jailbroke the apps into recommendations that largely matched. Every Grok API request also carried about 1,240 tokens of extra input the researchers did not write and could not see.

The models even searched differently before deciding: Grok surfaced a Christian conservative voter guide in 39% of answers and X in 41%, while GPT leaned on Wikipedia, the Associated Press, Axios and Reddit. With search switched off, GPT's lean stayed in the weights while Grok's vanished — it refused to pick in 27 of 30 races.