Chatbots show political bias and steer voters toward some parties, analysis finds

A new analysis from the University of Copenhagen has revealed that leading AI chatbots, including OpenAI's ChatGPT and Google's Gemini, exhibit significant political bias and are not neutral arbiters of information. When prompted for advice on who to vote for, the models systematically favored certain political parties over others, demonstrating that their training data and algorithms embed specific ideological leanings. The researchers conducted systematic testing across multiple political contexts and found consistent patterns of steering, raising serious questions about the use of these tools in democratic processes.

The study's authors explicitly warn that these AI systems are currently unsuitable for providing advice in connection with elections. The inherent bias makes them unreliable sources for voters seeking neutral information, potentially influencing electoral outcomes. This finding adds to growing concerns about AI's role in society, particularly as tech companies increasingly position these chatbots as general-purpose information tools. The research underscores the urgent need for greater transparency in AI training data and more robust bias mitigation techniques before such models can be trusted with politically sensitive queries.