Alert! Crafting evidence-based warning messages with Generative AI
Many emergency managers have seen or sent alerts like the below that missed the mark: acronyms that mean nothing to most people, unclear protective actions, and jargon that wastes precious characters while confusing recipients.
RCEMA ALERT 08:15 09/17/25: LVL 3 HAZMAT INCIDENT following train derailment at MP 28.3 BNSF corridor. Active release of Anhydrous Ammonia (NH3) confirmed. ERZ established for sectors 7A & 7B, generally N of Highway 54 & W of I-49. All persons in ERZ must initiate SIP protocols immediately. DO NOT ATTEMPT TO EVACUATE. Monitor EAS and local media for further instruction.
Alert sent by simulation county, Riverside County, Missouri, 2025.
In a dire situation, this kind of alert could cost lives. However, alerts like these are not outliers – they are symptoms of a systemic challenge.
Emergency managers know the scramble: a disaster hits and suddenly alerts need to go out across a dozen different platforms – each with its own quirks, limits, and rules. In the middle of the chaos, staff are juggling the requirements of the Integrated Public Alert and Warning System, (IPAWS): 90-character wireless emergency alerts (WEA), 360-character extended WEA alerts, broadcast Emergency Alert System (EAS) messages, commercial mass messaging platforms, and social media posts. One small formatting slip and a message that will reach thousands can cause more harm than good.
(...)
Although emergency management professionals are often initially skeptical of AI, their willingness to adopt it grows significantly with hands-on experience. Adoption offers clear benefits, such as greater efficiency, clearer messaging, and real-time data integration. However, concerns about accuracy, context, and cultural sensitivity persist.
(...)
Too often, studies on what makes warnings effective never make it into the tools emergency managers use. At the same time, technically solid systems can fall flat in the real world if they ignore how people behave under stress or if lessons from research never reach practitioners. By embedding all three perspectives in the development process, evidence-based practices became core features rather than afterthoughts.