Trust under threat: how AI vs. human mistakes in disaster alerts shape public perception and response

Abstract

As artificial intelligence (AI) becomes increasingly integrated into disaster communication systems, understanding public responses to AI-led alerts is critical for designing trustworthy and effective crisis messaging. In line with an emerging line of inquiry, this study represents early empirical efforts to examine how the public responds to disaster alerts generated and delivered by AI, addressing a critical and timely gap in the disaster communication literature. Across two experiments (N = 599), we investigate how the attribution of disaster alerts (AI vs. human expert) shapes perceptions of message credibility, trust, and disaster coping outcomes. Study 1 finds that AI-led alerts are perceived as significantly less credible than expert-led alerts, although no significant differences emerge in coping outcomes such as self-efficacy, collective efficacy, or behavioral intentions. Study 2 extends these findings by introducing a factual error in the alert, revealing that AI sources suffer greater credibility and trust losses than human sources, despite being blamed less. Across both conditions, participants held human actors, such as experts, developers, and deploying organizations, more responsible than AI, yet consistently rated AI as less credible and trustworthy, even when greater responsibility predicted lower trust. Notably, errors made by human experts also diminished future interest in AI-based alerts, suggesting an interesting and complex shift in cross-source trust. These findings offer new insights into AI as a social actor within the disaster communication ecology and highlight the need for transparent, human-in-the-loop designs when deploying AI in public crisis messaging.