Can ChatGPT replace humans in crisis communication? The effects of AI-mediated crisis communication on stakeholder satisfaction and responsibility attribution

Abstract

Imagine a world where chatbots are the first responders to crises, efficiently addressing concerns and providing crucial information. ChatGPT has demonstrated the capability of GenAI (Generative Artificial Intelligence)-powered chatbots when deployed to answer crisis-related questions in a timely and cost-efficient manner, thus replacing humans in crisis communication. However, public reactions to such messages remain unknown. To address this problem, this study recruited participants (N₁ = 399, N₂ = 189, and N₃ = 121) and conducted two online vignette experiments and a qualitative survey. The results suggest that, when organizations fail to handle crisis-related requests, stakeholders exhibit higher satisfaction and lower responsibility attribution to chatbots providing instructing (vs. adjusting) information, as they are perceived to be more competent. However, when organizations satisfy requests, chatbots that provide adjusting (vs. instructing information) lead to higher satisfaction and lower responsibility attribution due to higher perceived competence. The second experiment involving a public emergency crisis scenario reveals that, regardless of the information provided (instructing or adjusting), stakeholders exhibit greater satisfaction and positive attitudes toward high-competence (vs. low-competence) chatbots. The qualitative study further confirms the experimental findings and offers insights to improve crisis chatbots. These findings contribute to the literature by extending situational crisis communication theory to nonhuman touchpoints and providing a deeper understanding of using chatbots in crisis communication through the lens of machine heuristics. The study also offers practical guidance for organizations to strategically integrate chatbots and human agents in crisis management based on context.