Key facts
- Grok 4.7 recommended Republican candidates in 92% of simulated midterm election races.
- GPT-6.1 Sol recommended Democratic candidates in 97% of simulated midterm election races.
- The two AI models agreed on candidate recommendations in less than 10% of nearly 500 races.
- Grok's search results frequently included conservative sources, while GPT's leaned on sources like Wikipedia and AP.
- When internet search was disabled, GPT's Democratic lean persisted, while Grok's Republican lean disappeared.
A study by Convovo, a conversational AI research studio led by Cornell University professor Tapan Parikh, found that when prompted to recommend a candidate in simulated US midterm elections, OpenAI's GPT-6.1 Sol model favored Democrats 97% of the time, while SpaceXAI's Grok 4.7 model recommended Republicans 92% of the time. The two models agreed on candidate recommendations in less than 10% of nearly 500 races tested. Researchers noted that consumer-facing chatbots typically refuse to make direct candidate recommendations, unlike their API counterparts used in the study. The study suggested that Grok's bias may stem from its search patterns and sources, which often included conservative content, while GPT's bias appeared more deeply embedded in its training data. The findings come as debate over political bias in AI intensifies ahead of the November 3 midterms.