Recent research indicates that AI models exhibit distinct political leanings when asked to predict election outcomes. In a study covering nearly 500 races, xAI’s Grok showed a preference for Republican candidates, while OpenAI’s GPT favored Democratic candidates. The models chose the same candidate in less than 10% of the tested cases.
A separate study by researchers from Chalmers University of Technology and the University of Gothenburg analyzed six models: Claude, DeepSeek, Gemini, Mistral, GPT, and Grok. The team evaluated 148,302 responses based on 24,717 prompts related to 107 policy issues for the September 2026 Swedish parliamentary election.
The second study found that most models remained politically neutral, with GPT producing the most balanced outputs. Grok was identified as an outlier, showing a shift toward right-leaning positions on issues such as migration, crime, and gender. However, no model displayed a statistically significant bias toward any of the eight Swedish parliamentary parties.
