Customized GPT for Hazard Mitigation Planning: A Case Study on Usability and Implementation

• In Event: Proactive Responses and Historical Accounts of Disasters

Abstract

In the United States, the Federal Emergency Management Agency (FEMA) requires local jurisdictions to develop Hazard Mitigation Plans (HMPs) to be eligible for certain disaster assistance funds. These funds are a critical resource for communities seeking to rebuild and strengthen their resilience against hazardous events. However, these plans are often lengthy, generic, and lack actionable detail that can reflect community conditions, limiting their effectiveness for local implementation. In response to these challenges, stakeholders are exploring new approaches to improve the HMP process, including the potential of generative Artificial Intelligence (AI) to support and enhance hazard mitigation planning.

This study investigates the utility of a customized Generative Pre-trained Transformer (GPT) AI tool, "Hazard Helper", designed to aid in hazard mitigation planning. The central research questions are: (1) How do emergency management professionals perceive and engage with an AI assistive tool designed to support hazard mitigation planning? and (2) What insights does their interaction with the tool provide about their planning needs and challenges? The study involves two phases: (1) development of the Hazard Helper GPT tool trained on 200 county-level HMPs evaluated using a rubric developed by the research team and informed by FEMA standards, and (2) a pilot workshop with 30 emergency managers from the DMV (DC-Maryland-Virginia) region using Hazard Helper in four structured activities: plan summarization, follow-up queries, revision suggestions, and updates.  

Across the four activities, responses to the tool ranged from helpful to inadequate, highlighting the importance of human oversight. Participants found the tool useful in structuring and organizing information, especially for non-expert audiences and training purposes, and noted the custom GPT was helpful with generating useful visuals and tabular outputs. Nevertheless, the participants highlighted the need for more human validation, clear source citation, and improvements in the AI’s interpretive capabilities. This study points to the potential of AI as a complementary planning tool, emphasizing the importance of continued development grounded in expert input and ethical standards. Future research should prioritize improving model accuracy, incorporating more diverse stakeholder feedback, and integrating explainable AI features to enhance trust and usability in emergency management contexts. The paper offers contributions to theories of technology adoption and implementation and a process for co design of new AI systems with professionals in their field of expertise.

Author

• Muhammad Awfa Islam, Virginia Polytechnic Institute & State University
  Presenting Author