Modality and cultural tailoring in human–AI disaster communication: examining mechanisms and outcomes for Hispanic communities

Abstract

Purpose This study examines how GenAI chatbots can support Hurricane preparedness among Hispanic/Latino communities by comparing combined audio-text delivery with text-only delivery and culturally tailored with generic content. It focuses on information overload and emotional attachment as cognitive and affective mechanisms. Design/methodology/approach We conducted a 2 × 2 between-subjects online experiment with 186 Hispanic/Latino participants in Texas. Participants interacted with GPT-4o-powered chatbots that varied in (1) information modality (combined audio-text vs. text-only) and (2) cultural tailoring (tailored vs. generic). We also examined Hispanic/Latino acculturation as a boundary condition that may moderate the effects of modality and cultural tailoring. Findings GenAI chatbots combining audio and textual delivery increased information overload and decreased information-seeking intentions compared to text-only chatbots, and this modality effect on overload was stronger among participants with lower Hispanic/Latino acculturation. Cultural tailoring improved disaster preparedness and seeking intentions by fostering emotional attachment. This suggests that while cultural tailoring drives affective engagement, combining AI-generated speech and text may create cross-modal redundancy in disaster communication. Research limitations/implications This study extends limited capacity and cross-modal redundancy perspectives to GenAI disaster communication by showing how combined audio-text delivery can shape disaster outcomes through information overload, with Hispanic/Latino acculturation serving as a boundary condition. It also advances cultural tailoring research by showing how culturally tailored content shapes outcomes through emotional attachment. Practical implications Practically, public agencies should design modality-sensitive GenAI chatbots rather than defaulting to simultaneous audio-text delivery. Audio should be optional and user-controlled, with clear opt-in choices, replay, pause and speed controls. When audio and text are combined, the content should avoid unnecessary duplication and support pacing to reduce overload. At the same time, culturally tailored content can make disaster guidance feel more relevant and emotionally resonant for underserved Hispanic/Latino communities. Social implications As GenAI chatbots become more common in disaster preparedness, adding modalities by default may increase overload and reduce information seeking for some users, potentially widening resilience gaps. These findings underscore the need for culturally inclusive and cognitively accessible chatbot design. Public-facing systems should provide user control over modality and pacing, avoid redundant audio-text delivery, and account for within-group differences such as acculturation. Such design choices can help ensure that GenAI preparedness tools support, rather than burden, underserved communities. Originality/value This study examines an integrated model explaining how GenAI chatbot design shapes disaster outcomes through cognitive and affective mechanisms. It extends limited capacity and cross-modal redundancy perspectives by linking combined audio-text delivery to information overload, with Hispanic/Latino acculturation as a boundary condition, and advances cultural tailoring research by identifying emotional attachment as an affective mechanism. Peer review The peer review history for this article is available at: Link to the website