MIT Launches Dashboard to Track How AI Chatbots Answer Midterm Election Questions

How the dashboard works

According to MIT News, the team designed the dashboard to simulate how individuals across the country communicate with chatbots. Researchers put the same fixed set of questions to different AI systems. The topics include election certification, hot-button issues, result predictions and news coverage.

Every few days, the team runs the questions through the chatbots again. This lets them monitor how responses vary as the election gets closer, depending on user location, political party affiliation, identity and other factors.

The researchers say the aim goes beyond understanding how AI systems spread information. They also want to better comprehend how chatbots are shaping political discourse and voter understanding of political issues across the country.

Why a historical record matters

Podimata, who is an assistant professor of operations research and statistics at the MIT Sloan School of Management, said people are increasingly using AI chatbots to find and make sense of political and election-related information. She pointed to several complications: the models are continually updated, the political information available to them changes, and their answers can depend on how a user asks a question or what the user reveals about themselves.

One basic problem, she said, is that there is no good historical record of this information environment. If a chatbot gives one answer today and a different one weeks later, it is very difficult to reconstruct afterward what voters saw at each point. One goal of the project is to document that environment systematically and make the data available to other researchers and the public.

Podimata also said factual accuracy is only the first frontier. Differences in framing, emphasis, uncertainty and cited sources can matter enormously, she said, even without outright misinformation.

Building on 2024 research

The observatory builds on Podimata's earlier work during the 2024 presidential election. According to her, that research found meaningful variation in AI models' responses over time and across prompt framings, as well as differences in how models characterized candidates and implicitly anticipated the election outcome.

Midterms offer a chance to expand that work, she said, because many state and local contests unfold at the same time as broader national narratives. The new audit survey therefore includes contest-specific questions about U.S. House and Senate candidates, as well as broader questions about voting information and other aspects of the election information environment.

Invisible differences between users

Podimata argued that people build their own "microcosms" through their interactions with chatbots. Someone may not realize that their favorite chatbot says something quite different about a political issue than a friend's favorite chatbot does. Chatbots may frame the same facts differently, emphasize different aspects, express different levels of uncertainty or draw on different sources, even for identical queries. Two users could come away with very different perceptions of the same facts without knowing it.

She compared this with social media, where users and researchers could at least partly observe what circulated in Facebook groups, Instagram accounts or TikTok communities. With chatbots, she said, these environments are largely invisible to anyone outside the individual conversation.

Questions the data should answer

Podimata listed several questions the dataset is meant to help study:

• Do different models provide systematically different information for the same questions?
• Do their responses change over the course of the campaign?
• Does a model's answer change depending on how a user frames the question or describes themselves, for example as a Democrat or a Republican?
• Which sources do the models draw on when answering election-related questions?

The public dashboard is an important part of the project, she said. The team wants people inside and outside academia to see the data as the election unfolds, rather than only seeing results months later.