The promise and limitations of using GenAI to reduce climate skepticism

Two experiments examined whether dialogues with Generative AI can effectively reduce climate skepticism. In Study 1 (N=949), climate skeptics who engaged in short conversations with ChatGPT showed small but significant increases in pro-environmental action intentions and decreased confidence in their initial skeptical climate views. In Study 2 (N=333), similar effects were observed and extended to modest improvements in climate skepticism and policy support. However, these effects were small and, in the case of skepticism, inconsistent across studies and not durable over time. Increasing the conversation length from 3 to 6 rounds yielded no additional benefit. Generative AI shows promise as an accessible and scalable educational tool, particularly because it can deliver personalised information while circumventing some of the source-based reactance common in politicised contexts. However, effects were modest, variable across studies, and susceptible to decay, underscoring the need for cautious, evidence-based integration of Large Language Models into climate communication strategies.