Q&A: Hey AI, who should I vote for this year?

MIT researchers have developed a new tool called the LLM Election Observatory to track how artificial intelligence chatbots respond to questions about the upcoming 2026 U.S. midterm elections. Led by Professor Adam Berinsky, Assistant Professor Chara Podimata, and Professor Charles Stewart III, the team aims to understand the role AI plays in shaping political discourse and voter understanding.

The LLM Election Observatory simulates real-world interactions with chatbots, monitoring how different systems answer fixed sets of questions across various topics, including election certification, hot-button issues, result predictions, and news coverage. The dashboard considers factors like user location, political party affiliation, and identity to analyze the responses.

One challenge in this research is that AI models constantly update, and their responses can change depending on the user's question and self-provided information. Moreover, there is no historical record of the evolving information environment surrounding elections, making it difficult to reconstruct how voters saw information at different points in time.

Beyond factual accuracy, the project aims to explore how AI chatbots may subtly influence voters' understanding of political issues. Differences in framing, emphasis, uncertainty, and sources can have significant impacts on how people perceive political matters, even without spreading misinformation. For instance, users might unknowingly receive different information from chatbots used by friends or family members, leading to varying perceptions of the same facts.

Podimata and Berinsky emphasize the need to comprehend how individualized information environments are shaped by users' interactions with chatbots and how they evolve over time. This issue mirrors concerns in social media, where users and researchers can observe information circulating in different online communities. However, with chatbots, these information environments remain largely invisible to outsiders, making it challenging to monitor and understand their effects on voters.