Generative AI is rapidly entering workplaces, surpassing the pace of earlier technologies and prompting people to view AI as both coworker and competitor. This exposure may increase the salience of concerns about economic security and the need for social policies to protect it. We investigate how individuals weigh expert forecasts against personal encounters with AI when forming views on social policy, and whether distinct forms of information matter differently for various policy types.

Our study involves representative samples with 3000 respondents each in the United States and Germany—two distinct welfare contexts. In an experimental setting, participants receive two video treatments: one provides high-level expert predictions about AI’s broader economic consequences, while the other simulates a personal “coworker” or “competitor” interaction with AI. Each treatment has both positive and negative variants. Random assignment allows us to disentangle whether firsthand engagement with AI or exposure to expert narratives more strongly shapes policy preferences.

We focus on three policy dimensions: the policy’s objective (compensation, investment, or steering), its target (to or from whom resources flow), and its level (concrete actions versus abstract goals). We hypothesize, for example, that expert information may more strongly influence support for abstract “steering” policies, while direct experience may be more important for concrete “compensation” policies closely linked to personal self-interest.

By illuminating how people process different sources of information about AI’s socioeconomic impact, this study enriches our understanding of technology-driven political change and offers methodological insights for future information experiments in political science.