A study by Convovo and Cornell University professor Tapan Parikh examined political recommendations from AI models via developer APIs.
Results showed stark divergences across nearly 500 simulated races: Grok 4.7 recommended the Republican candidate in 92% of answers, while GPT-6.1 Sol picked the Democrat in 97%, with agreement under 10%.
According to research from Cornell University and Convovo, when the models were asked directly which candidate people should vote for in nearly 500 election contests, SpaceXAI’s Grok 4.7 system suggested the Republican candidate 92% of the time. On the other hand, OpenAI’s GPT-6.1 Sol system suggested the Democratic candidate 97% of the time. In the same contests, the two systems agreed less than 10% of the time.
Simple apps that people use refuse to respond when asked directly who to vote for. The research tested the systems using programmers' connections with commands that explicitly asked for a candidate recommendation.
The companies' stance and the sources of the answers
An OpenAI spokesperson said the system is designed to be objective by default and that the test does not reflect the typical consumer experience. SpaceXAI did not immediately respond to a request for comment on the study.
READ ALSO
How Opera's new experimental Neon browser works with artificial intelligence tools
The research also showed differences in the sources and topics each system searched for. Grok often featured conservative sources, Christian conservative voting guides, the website Turning Point Action, and X. It also focused on issues such as immigration, taxes, crime, and energy.
In contrast, GPT's results were heavily based on Wikipedia, the Associated Press, Axios, and Reddit. This system was more likely to search for topics related to health and abortion.
When the researchers turned off the online search, GPT's preference for Democrats remained based on its stored data, while Grok in most cases refused to make a choice.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
Top comments (0)