In an experiment conducted with 2346 US residents, we examine the effect of content (positive/ negative/ inconclusive) and style (scientific/unscientific) of narratives about \(\textrm{CO}_2\) emissions on belief updating and pro-environmental behavior. We use the Work for Environmental Protection Task (WEPT) by Lange and Dewitte (Behave Res Methods 54:133–145, 2022) as a proxy for pro-environmental behavior. Narratives are constructed using the natural language processing chatbot ChatGPT. We find that negative narratives significantly increase concern regarding \(\textrm{CO}_2\) emissions, while positive narratives significantly decrease concern, with the latter effect being significantly larger in terms of absolute value. For inconclusive narratives, the effect on beliefs depends on prior beliefs: Subjects with low prior beliefs increase their concern, whereas subjects with high prior beliefs decrease their concern. Moreover, we find that the unscientific style has a stronger effect on belief updating than the scientific style. Neither content nor style affects pro-environmental behavior significantly.
