Deep conversations felt less awkward, and produced more connection and more happiness, than the people having them expected. The mis-prediction was largest for exactly the deeper topics — people are most wrong about the conversations that matter most.
How it was done
Experiments assigning people to have either shallow or deep conversations, measuring predicted awkwardness and connection beforehand against reported experience afterwards.
What it means
People systematically over-predict awkwardness and under-predict connection, and they do it more as the topic gets more substantive. So the bias is worst precisely where the cost of avoidance is highest.
One proposed mechanism: people underestimate how interested others are in them, so they expect a deep disclosure to land in indifference. It generally does not.
This is the same shape as the reaching-out finding. The forecast is wrong; the experience is better than the forecast.