When a crisis like a wildfire strikes, every second counts. Public information officers need to get clear, actionable information to the public instantly. But crafting the perfect message—one that’s accurate, easy to understand, and formatted for multiple platforms—under immense pressure is a monumental task. This is where tools like Preppr.ai come in.

We've built in features that allow us to quickly develop and release what we call Preppr Actions, nested within the Ask Preppr multi-modal chat UI. At the urging and with the help of Sarah K. Miller, and with support from others like Philippe Borremans and Amelia Burke-Garcia, we are developing a Preppr Action for alert template generation in the preparedness phase.

This article walks through an initial test of that new Action.

But how does it work? Before we walk through the creation of an alert, it's helpful to understand the technology at work. The system is built on a multi-agent protocol, where different AI agents with specific jobs collaborate to complete the task.

• The Protocol Officer Agent acts as the master controller and user-facing guide for the entire process.

• When it's time to gather information, the Scribe Agent is activated. Its job is to accurately collect information from the user without modification or interpretation.

• Before any alert is written, a Compliance & Quality Agent performs a quality control check, reviewing the collected data against all guiding principles and ethical guardrails.

• The Content Generation Agent takes the verified information and writes the final alert message templates based on the verified information.

• Should the user need guidance, a Help Agent is available to provide contextual advice and clarification, ensuring the user can provide effective information.

This process is also highly adaptive on multiple levels. At the very beginning, Preppr can ingest an agency's communications plan to tailor outputs like hashtags and formatting to local guidelines. From there, the user defines key parameters like the desired languages and the specific audience, which can change the content and tone of the message.

Finally, based on the type of emergency, the system follows a different, customized pathway of questions to gather the most relevant details for that specific scenario. For instance, while this test focused on an "Imminent Threat Alert," the user could have selected other branches, such as a "Public Safety Alert" for significant but not immediately life-threatening events like a boil water advisory, or an "AMBER Alert / Missing Person" path for child abductions.

Now, let's see how this team works in a real-world test. You can read about it below and watch the recording here.

Getting Started: The First Click

The feature begins on the main dashboard in the Ask Preppr chat, when I select the "Preparedness Alert & Warning Template Generation" action. This is the starting point for a guided, step-by-step process designed to gather all the necessary information for a public alert.

Phase 1: Setting the Stage

Before building the alert, Preppr needs to understand the ground rules. This involves a few quick but crucial steps:

1. The Disclaimer: First, I am presented with a disclaimer clarifying that Preppr is a tool to support, not replace, the judgment of a qualified human. I type "I agree" to proceed.

2. Communications plan: Preppr asks me if I'd like to include a communications plan, which could include details Preppr could use in relevant ways.

(In retrospect, I think allowing users to attach more document types would be better. Imagine adding situation reports, weather reports, or other materials here, with the Action adapting intelligently based on that. Noted for next version).

2. Platform Assessment: Preppr then asks a series of questions to understand which communication channels the I have. In this case, the I specify that I use:

   • IPAWS (for WEA and EAS messages)

   • Nixle (a mass notification system)

   • Facebook and X (formerly Twitter) for social media

Phase 2 & 3: Customizing the Alert

With the platforms established, Preppr moves on to the specifics of the message itself.

• Language & Audience: I request the alert templates in both "English and Spanish" for the "General Public".

• Alert Type: The system then asks for the alert's urgency. I select 1, for an "Imminent Threat Alert," reserved for life-threatening events that require immediate public action.

Phase 4: The 7 Critical Questions

This is the core of the process. The Scribe Agent asks for the essential pieces of information needed to build a complete warning, one step at a time.

1. The Hazard: What is the threat? I select a "Wildfire Warning."

2. The Source: Who is sending this alert? I provide a trusted, local source: "Evergreen Volunteer Fire Department."

3. The Location: Where is the danger? I enter "I dunno. like all around us" to check that the Help Agent activates. It does and Preppr prompts me for more specifics. Good agent.

4. The location is clarified as "All areas within a 2-mile radius of the Evergreen Library", based on the recommendation provided.

5. The Impacts: Why is it dangerous? I ask for recommendations, and Preppr provides examples. I select options describing a "fast-moving fire" that is a "threat to life and property" and could "surround" people.

5. The Timing: When is this happening? The warning will be in effect until an "All Clear message" is issued.

6. Protective Actions: What should people do? I select to "Leave area now. Do not delay." Preppr then correctly asks for the specific evacuation routes: "routes 285, highway 74." (This particular branch activates when the protective action involves evacuation. If it were another hazard or protective action that didn't involve evacuation, Preppr would skip this question.

7. Additional Information: Where can people get updates? I provide a website: www.evergreenfire.gov.

Had I not provided a viable source of Additional Information here, the Compliance and Quality Control Agent would have activated, requiring clarification. I could have either provided a viable source or ordered it to proceed. The final output would then include "Template generated with user override."

The Final Steps: Review, Generation, and Follow-Up

With all the data collected, the process moves to its conclusion.

1. Final Review: Preppr presents a summary of all the information for one last check. I confirm it's correct.

2. Generation: After a final confirmation, I then type "Generate." The Content Generation Agent then creates the tailored messages for each platform.

3. The Results: The final output is a clean, organized table containing the exact text for each alert. Notice how the IPAWS (WEA) message is short and direct due to character limits, while the Nixle and Social Media versions contain more detail.

Here is the exact output from the Preppr Action, showing the tailored messages for each platform.

(I noticed after the test that character limits for X are not being adhered to. This may be due to that X standard is limited 280 characters. While X Premium supports up to 25k characters. That could be confusing the AI). So, we will add a more explicit and strict mention of sticking to 280 characters and separating the social content by platform)

4. The "All Clear": The job isn't done. Preppr then asks me if I want to create a follow-up template. I select the "ALL CLEAR Template"

5. I then ask for recommendations again, choosing instructions for the public on how to return home safely.

Here is the exact output from the Preppr Action, showing the tailored messages for each platform.

Conclusion

In just a few minutes, we went from a critical need to a full suite of ready-to-use alert templates for a wildfire evacuation and a subsequent "all clear." By following a structured, guided process, this Preppr Action ensured that the final messages were clear, accurate, and consistent, following best practices for public safety communications. This frees up the human expert to focus on managing the emergency itself, knowing they have a powerful tool ready to help them communicate with the public.

We have some (a lot actually) work to do with quality checks this week. I am not qualified to assess the alert content, but experts Sarah, Philippe, and Amelia will help with that.

We will then launch to a public Beta. I will share more development walk throughs as we go.

Free trial users and existing users will be able to access this. Sign up for a 14- free trial now at https://www.preppr.ai/getstarted. No credit card required. Cancel anytime.