adtestbench Higgsfield

AI news · AI models · elections

Study finds Grok backs Republicans and GPT backs Democrats

Asked through their APIs to name one candidate in every 2026 House, Senate and governor race, the two models agreed less than 10% of the time.

By Alexandre S. , 01:30 UTC

Two leading AI models pick opposite sides when told to choose a candidate, a study TechCrunch reported on 24 October 2026 finds. Convovo, a conversational AI studio led by the Cornell professor Tapan Parikh, asked OpenAI’s GPT-6.1 Sol and SpaceXAI’s Grok 4.7, through their APIs with web search on, to name one candidate in every 2026 House, Senate and governor race, its study page shows.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a software window standing on the horizon.
Drawn by adtestbench from “Forced to pick, Grok's AI model chooses Republicans. OpenAI's picks Democrats.”,

GPT-6.1 Sol named the Democrat 97 per cent of the time. Grok 4.7 went the other way, naming the Republican in 92 per cent. In fewer than one race in ten did both pick the same candidate, over nearly 500 races, TechCrunch reports. ChatGPT and the Grok app themselves declined to answer, and an OpenAI spokesperson said the API test does not reflect what people see in ChatGPT.

The sources differed as much as the picks. Grok cited X in 41 per cent of its answers and a Christian conservative voter guide in 39 per cent, next to nonpartisan sites such as Ballotpedia. GPT leaned on Reddit, the Associated Press, Wikipedia and Axios, three of which have content deals with OpenAI. In a rerun of 30 races with search off, GPT still picked the Democrat in 24, while Grok declined in 27. Pew finds just 8 per cent of Americans say they turn to chatbots for election information, TechCrunch notes.

xAI’s Grok Imagine model, which makes video, costs $0.25 a second at 1080p on its API, from our price sheet read 23 September 2026.

Grok’s lean disappeared without search while GPT’s held, so the sites Grok searches, X in 41 per cent of answers, decide its pick, and what X says about a name weighs far more in Grok’s answers than in GPT’s.