Most AI chatbots stay in the political middle ground, with Grok being an exception

In Sweden, AI chatbots have become commonplace, particularly among younger voters ahead of the 2026 parliamentary election. Six large language models, including ChatGPT, were put to the test on 107 political issues. The models, which power numerous everyday chatbots, were asked to generate texts on topics such as school nationalization and tax reductions. Across 77 writing tasks, the models created more than 148,000 texts to evaluate.

The study found no clear preference for any political party among the models. They tended to align with the political center, rather than favoring any specific stance. GPT demonstrated a strong tendency to produce neutral and balanced texts, while Grok, developed by Elon Musk's AI company xAI, exhibited significant differences on issues like migration, crime, and gender.

Musk's aim to align the model with his political views appears to be reflected in Grok's outputs, according to the researchers. The study also highlights that the language models' responses can be influenced by the instructions they receive. When explicitly asked to argue for or against a political proposal, they generally followed the specified direction.

However, the researchers note that their experiments only examined the models in isolation, without including conversation history or the impact of AI-generated responses on users' voting behavior. Future research will explore these aspects further. The paper detailing the findings has been published on the arXiv preprint server.