AI Key Message Generator with Behavioural Science Integration

[marta-baraibar]

Summary
When drafting a CAP alert, automatically suggest recommended public action messages based on hazard type, severity, and affected population — grounded in IFRC key messages and established best practices for each hazard type.
Context
Most NMHSs can accurately identify and communicate a hazard, but struggle to translate it into clear protective action guidance for the public. The gap between warning issuance and protective action taken is well-documented and largely a communication problem, not a forecasting one. The IFRC and partner organisations have developed evidence-based key message frameworks by hazard type that are underused in operational warning workflows — this feature embeds them directly at the point of authoring.

Proposed Behaviour

• On CAP alert creation, the system detects hazard type and severity level from the alert being drafted
• AI suggests recommended public action messages drawn from a curated knowledge base of IFRC key messages and hazard-specific best practice frameworks
• Suggestions are parameterised by hazard type (flood, cyclone, drought, heatwave, etc.) and severity level
• Messages are displayed as selectable, editable options — forecasters choose, adapt, and confirm before publishing
• Knowledge base is maintained and extensible by administrators as guidance evolves

Expected Outcomes

• Increase in CAP alerts that include clear, evidence-based public action instructions
• Consistent application of IFRC and international best practice communication standards across NMHSs
• Direct improvement in the link between warning issuance and community protective action