Human versus AI-generated emergency alerts: a comparative study for AI emergency communication applications

This study explores the capabilities of artificial intelligence (AI) to generate emergency alerts, using Tropical Cyclone Kirrily as a case study, which made landfall in Queensland, Australia, on 25 January 2024. ChatGPT-4 was used to recreate official alerts issued during the event via structured prompts informed by the Internalisation, Distribution, Explanation, Action (IDEA) model and Australian disaster communication guidelines. Findings indicate that AI-generated alerts largely satisfied the core IDEA model components and demonstrated a strong ability to personalise messages for specific needs and audiences with clear safety instructions. However, human oversight remained necessary, particularly as conditions escalated, to ensure alerts reflected the evolving risk context. The comparative analysis undertaken shows that while both human- and AI-generated alerts had limitations, AI has the potential to enhance clarity and personalisation of emergency messages. Overall, the study highlights the value of combining structured risk communication frameworks with human oversight to effectively integrate AI into emergency communications.