Large-scale listening: the most underrated approach to building a campaign message
The political industry is using AI to test ideas a range of ideas that may not teach us anything about how voters are really making decisions. According to the latest episode of the Altum Insight Podcast, these applications may be downstream of the most impactful changes we could make in terms of research and message development.
Co-hosts Adrianne Marsh and resident mad scientist Frank A. Spring sat down to discuss how we can think differently about the potential role of AI in winning elections. The takeaway? The best use of AI in politics isn't content generation, it’s listening. As Spring says:
"If the AI is making things easier all the time and not making things harder and not provoking strategic conversations that are difficult, you're using it wrong."
Traditional polling tells us what people think, but it completely fails at explaining why. To fix this, Altum uses AIMI (AI Moderated Interviewer)—an AI agent capable of conducting dynamic, open-ended audio interviews with thousands of voters at once.
Instead of forcing a voter into a scripted conversation, AIMI asks an open question and builds custom follow-ups on the fly. Surprisingly, behavioral research shows voters are often more honest with an algorithm than a live human operator because they don't feel judged. Spring explains:
"In of itself, that is a remarkable tool because it allows you to harvest qualitative research, to ask big open ended questions at scale, and to make sense of the results, which is not something that was ever really possible before."
AI is good at processing lots of complex voter language quickly, but it will only help us understand the throughlines of this unsettled electorate if we ask the right questions and interrogate the narrative undercurrents beneath what participants are sharing.
Drop in to hear how Altum is exploring what it means to listen differently. Episode 4 is below as well as on Spotify and Apple.
Listen on Spotify:
Listen on Apple Podcasts:
