Framing artificial intelligence: risk and benefit narratives in a preregistered experiment in Denmark

Abstract Public communication about artificial intelligence (AI) alternates between narratives of benefit and narratives of risk, but evidence on whether these frames move public opinion remains limited. We conducted a preregistered, three-arm online survey experiment in Denmark using a quota-based sample from a national panel. Respondents were randomly assigned to a neutral definition of AI, a risk-focused message paired with a cautionary image, or a benefit-focused message paired with an optimistic image. We measured overall evaluations of AI’s societal impact and agreement that AI erodes critical skills such as critical thinking and independent problem solving. We also elicited open-ended reflections on AI and job opportunities in respondents’ fields, as well as on generational dependence on AI, which we coded using a codebook-guided large language model workflow. We find that benefit framing increases positive overall evaluations relative to the neutral control, whereas the risk frame does not produce reliable change. In contrast, perceived skill erosion shows little movement across conditions, and open-ended reasoning about work and dependence exhibits similar thematic distributions across experimental arms. These results suggest that brief frames can nudge general sentiment about AI while leaving more specific capability concerns and immediate reasoning largely unchanged in a high-trust, highly digital welfare-state context.