Investigating the Impact of Chatbots versus Text-Based interventions on Environmental Awareness, Agreeness with Environmental information and credibility

Abstract

With recent improvements of LLMs, environment-related knowl- edge augmented conversational agents (CAs) have become feasible for increasing environmental awareness. However, their benefits and limitations alongside with how different valence levels of po- larizing information affect user trust need investigation. Particu- larly, how user agreement, reactance and environmental awareness differ between CA and static text in a long-term setting remains unclear. In this pre-registered study, we conducted an online multi- week mixed-method intervention (n = 449) with information format (CA, text) and valence (positive, negative, and neutral) as between- subject factors and time as a within-subject factor. Results indicated higher agreement and lower reactance in the CA condition com- pared to texts, regardless of their valence. Environmental awareness significantly increased over time especially for users with low level of envrionmental awareness but showed no difference by format or valence. Trust increased over time depending on the valence condition. We discuss CA design and methodological implications for future research.