Two strands. One on whether balanced coverage changes what people read, one on how people work with agentic AI.
The question is whether showing people left, centre, and right-leaning coverage of the same story side by side reduces affective polarisation. I built it as a field experiment rather than a survey, because stated appetite for the other side and actual appetite for it are different quantities and only one of them predicts anything.
That meant building a working newsletter, recruiting a real audience, running it against a single-perspective control, and measuring what people opened rather than what they told me they valued. The multi-perspective version drew a 69 percent higher open rate.
My part of the project is the analysis. Mixed methods in R and Excel, coding qualitative responses to a scheme consistently enough that a second coder would land in the same place, and checking that a quantitative result holds when the specification changes rather than reporting the first version that looked good.