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

MIT researchers have launched the LLM Election Observatory, a public dashboard that monitors how popular AI chatbots answer questions about the 2026 U.S. midterm elections, MIT News reported on October 30. The project is led by professors Adam Berinsky, Chara Podimata and Charles Stewart III. Every few days, the team reruns a fixed set of questions on topics such as election certification, hot-button issues, result predictions and news coverage, varying factors like user location, party affiliation and identity.

Podimata said the effort addresses a missing historical record: once a chatbot's answer changes, it is hard to reconstruct what voters saw earlier. She argued that accuracy is only the starting point, since framing, emphasis, uncertainty and cited sources can shape perceptions even without misinformation. Unlike social media, she noted, chatbot conversations are largely invisible to outsiders. The work extends her 2024 election research, which found meaningful variation in responses over time and across prompt framings.

Full longitudinal analysis has not been done, but Podimata said early 2026 data show differences in presentation, sources and tone across user identities, sometimes with the same underlying facts. She cautioned that the team cannot yet say how stable those differences are or how real people react to them.

Berinsky warned voters that a confident-sounding chatbot is not necessarily an accurate one, calling its answers a reasonable starting point only. He also said AI cuts both ways, able to mass-produce misleading content while showing promise in debunking conspiracy beliefs. The team plans to make its dataset available to other researchers.