A Practical AI Playbook for the 2025 Hurricane Season

Note from the President

The Atlantic hurricane season begins on June 1. We have eight weeks. April is the month, in this house, when the preparation actually happens, because by May the to-do list is too long and by June the rains are already starting. This year we have a longer to-do list than usual, because the early seasonal forecasts from Colorado State and the UK Met Office both call for an active season, and because the institutional memory of how badly the region was hit by Beryl last July is still fresh.

Dr. Marcus Charles returns to our pages this month with a practical piece on how AI tools should and should not be used in your hurricane preparation. The piece is concrete. Read it with a notebook open.

A second short item. April is Autism Awareness Month, and the Working Group on AI in Health is, this year, running a small consultation with parents and educators of autistic Caribbean children on what AI tools have been useful, what has been harmful, and what gaps remain. If you want to contribute, the address is health@caribbeanaiassociation.com. Earth Month is also this month. The two themes share more than you might think. The mind that protects itself from sensory overload and the region that protects itself from climate volatility are both, in the end, asking the same question of the technology in front of them. Will you make my life easier or harder. Build accordingly.

Adrian Dunkley Founder and President, Caribbean AI Association

---

Feature

A Practical AI Playbook for the 2025 Hurricane Season

By Dr. Marcus Charles

We have learned, painfully and repeatedly, that there is no single tool that solves a hurricane. There is a season of preparation, a window of warning, a period of impact, and a long recovery. AI tools can help in all four phases, and they can also waste your time and money if applied to the wrong phase or with the wrong assumptions. This piece is organized accordingly. Pick the section that fits where you are.

Phase one. Preparation. April through May.

The work in this phase is unglamorous. Update your inventory. Stock the cupboard. Service the generator. Trim the trees. Update the family communication plan. The AI tools that help here are the ones that help with general productivity, not the ones that promise hurricane-specific intelligence.

Three concrete applications.

First, a personal preparedness audit. Ask Claude or ChatGPT to walk you through the standard preparedness checklist for your specific household. Tell it your address, your household size, whether you have children, elderly relatives, pets, medical needs, a generator, solar, well water, septic, and so on. It will produce a tailored checklist faster and more thoroughly than you would on your own. Save the output. You will use it again in June.

Second, an insurance review. If you have a homeowner or business policy, feed your declaration page into a model, with your personal information redacted, and ask it three questions. What perils are covered. What are the exclusions. What is the deductible structure, particularly for named storms and hurricane sub-limits. Caribbean property policies frequently have separate, higher deductibles for hurricanes, which homeowners discover for the first time when they file. Knowing this in April beats knowing this in October.

Third, a small business continuity sketch. For business owners, ask a model to produce a one page continuity plan for your specific operation under three scenarios. Forty eight hour power loss. Two week communications loss. Six week site inaccessibility. The model will not produce a finished plan. It will produce a starting point that takes thirty minutes of your time and an hour of refinement, against the alternative of never writing one at all. Most Caribbean small businesses do not have a written continuity plan. After Beryl, the survivors were over-represented among the ones that did.

Phase two. Warning. The seventy two hours before landfall.

Here, the AI tools are most useful for information triage. A storm warning produces a flood of communications, official advisories, social media speculation, conflicting reports from neighbours and family. The work is to find the signal in the noise.

The official primary sources do not require AI. The National Hurricane Center advisories, your country's meteorological service bulletins, the CDEMA situation reports, and the local emergency management agency communications are the only sources whose timing and content you should plan your decisions around. Do not let an AI assistant become a substitute for these. Subscribe directly. Set the notifications.

Where AI does help, in the warning phase, is in translating and summarizing for your specific situation. A model can take the latest NHC discussion, which is written in technical language, and explain it for your particular geography, your particular elevation, your particular construction. It can compare the forecast track to a map of your specific coastline and tell you, with reasonable accuracy, what historical analogues most closely resemble what is coming. It can produce a tailored list of last minute preparations for the time you have left.

Three cautions in this phase. First, do not ask the model to forecast the storm. The frontier general purpose models are not weather models. They will hallucinate forecast tracks, intensities, and timing if you ask them to. Use them to interpret official forecasts, not to produce alternatives. Second, do not rely on the model's training cutoff for current information. If you are reading a model that has not been updated with the past forty eight hours of observations, its situational awareness is worse than yours. Tell it the current information. Do not ask it to know. Third, be careful with image generation. There has already been a pattern, after several recent storms internationally, of AI generated images being mistaken for real photographs and amplified by well-meaning local media. The damage from disinformation is real. Do not generate, do not forward without verifying, and treat any image too dramatic to be true as suspect until proven otherwise.

Phase three. Impact. The storm and the immediate aftermath.

During the storm itself, your AI tools are unlikely to help. Your power is out, your internet is out, your phone tower may be out. The work in this phase is the physical work of getting through it.

In the first hours after the impact, when communications begin to return, AI tools can do two specific useful things. They can help you draft and translate communications to family abroad, particularly across the languages of our region. They can help you produce structured damage reports for insurance purposes, which speed up claims processing by several weeks when done well. The model is not a substitute for a loss adjustor. It is a fast way to produce a structured, dated, photographically anchored inventory of damage in the form that insurance companies process most efficiently.

There is one application I want to flag specifically for our region. In areas where road damage and communication outages slow the arrival of formal needs assessments, structured AI assisted reporting from residents, channelled through CDEMA or CAIRA's emergency information node, has the potential to shorten the lag between impact and informed response by days. We are working with CDEMA on the design of this for the 2025 season. More on this in our June issue.

Phase four. Recovery. Weeks and months after.

This is where AI tools earn their keep, and where most Caribbean households and businesses are unaware of how much help is available.

For households. Claim documentation. Schedule management for contractors. Cost comparison for replacement materials. Drafting of correspondence with insurers, the government disaster relief authority, and the bank if mortgage payments need to be deferred. The work of recovery is largely paperwork, and paperwork is what these tools handle best.

For small businesses. The same plus several more. Customer communication during the period of disruption. Supplier renegotiation. Cash flow modelling under different recovery timelines. Application drafting for the small business grants that several Caribbean governments deploy after major storms but that most eligible businesses fail to apply for because the paperwork is daunting. After Beryl, three quarters of eligible Jamaican small businesses did not apply for the disaster grant programme they qualified for. The most common reason given was that the application was too complex. A model can produce the first draft of that application in twenty minutes from a conversation about your business. Whether you finish it and submit it is up to you.

For communities. The model in church basement, in community centre, in school hall, used by the local recovery committee to draft applications, summarize meetings, translate documents, and keep the long list of who needs what visible. The most effective post-storm recovery efforts I have observed in the region are the ones that combine deep local knowledge with reasonable administrative capacity. AI tools, on a basic free or low cost tier, are now an administrative capacity multiplier that the smallest community organization can access. CAIRA's Community Resilience programme is publishing a template kit of these for any registered community organization in a CAIRA member country, available from May.

A few tools, named.

For the household, ChatGPT Plus or Claude Pro at twenty US dollars per month each is the right tier. The free tiers are too rate limited during the high traffic days around a storm. Get a paid subscription in May. Cancel it in December if you want.

For the small business, the same plus Google Gemini Workspace at the business tier if your business already runs on Gmail and Google Drive, which most Caribbean small businesses do. The integration with your existing files is the value.

For the community organization, the free or non-profit tier of any of the above, plus the open NotebookLM from Google which is useful for working through long documents and which is, at the time of this writing, free.

A closing word.

I have spent twenty hurricane seasons in this region as a working professional, and seven as a parent. The lesson the seasons keep teaching is the same. The technology you have access to matters less than the planning you have done with it. The most prepared households I know would be prepared with a notebook and a wind-up radio. The least prepared would be unprepared with every tool on earth. AI is, this year, a useful new instrument in an old toolkit. The toolkit is still the toolkit. Pick up your tools. Use them well. Stay safe.

---

Dr. Marcus Charles consults on technology adoption with Caribbean businesses and serves on the CAIRA Community Resilience Working Group.

Originally published in The Monthly Intelligence Report, April 2025.