Practical AI work in Belize
Created by Adrian Dunkley, MaestrosAI, for the LAC AI Playbook.
These are proposed workflows for local businesses and public-service teams, not evidence of adoption, a national ranking or a forecast of economic returns. Choose one task with a checkable output and a responsible reviewer.
Before starting
Use synthetic or approved non-sensitive information. Do not upload identity documents, medical records, financial histories, customer or employee files, or children’s information to a general chatbot. Check the tool’s current eligibility, fees, retention and access controls before use.
AI drafts must not decide credit, immigration, employment or healthcare outcomes. Qualified professionals and responsible authorities retain control of legal, clinical, engineering, navigation and emergency decisions. Drafting does not authorise sending, filing, purchasing or publishing.
1. Visitor information for reef and inland operators
Prepare marketing and guest-information drafts from the operator’s actual services. Dive, cave, wildlife and other adventure safety decisions belong to qualified operators. Do not infer that a place name establishes visitor access, UNESCO status or permission to use a photograph.
Starter prompt
Use these approved operator sources to draft two visitor-information options. Keep prices, dates, accessibility limits and transport details exact. Do not generate dive briefings, cave routes, wildlife-handling instructions or safety assurances. Mark unconfirmed stops for staff review.
Check before use: Compare every consequential statement with the supplied source. Record corrections and require the responsible person to approve the result.
2. Farm and buyer records
A grower can organise delivery, quality and cost records for a discussion with buyers and an extension officer. Do not generate pesticide or fertiliser schedules or imply that a trade preference applies without checking the product and route.
Starter prompt
Create a worksheet from these synthetic or approved farm records. Show quantities, units, dates, buyer requirements and missing data. Leave input prescriptions and crop diagnosis to the agricultural adviser. Do not invent certification, future yields or an export arrangement.
Check before use: Compare every consequential statement with the supplied source. Record corrections and require the responsible person to approve the result.
3. Professional-services source checklist
A corporate-service team can organise current public requirements for local counsel to review. Do not recommend offshore structures, bank accounts or tax outcomes from model memory.
Starter prompt
From these current official documents, create a blank compliance-requirement register with sources and effective dates. Do not use client financial or identity records, assess eligibility or certify compliance. List questions for qualified local counsel.
Check before use: Compare every consequential statement with the supplied source. Record corrections and require the responsible person to approve the result.
4. Customer-support draft testing
A service or contact-centre team can test replies on synthetic enquiries against an approved FAQ. Do not assume a model will multiply ticket capacity or improve quality without measurement.
Starter prompt
Draft replies to these synthetic enquiries using only the approved FAQ. Flag unsupported answers and route complaints or commitments to a human. Measure total review time and incorrect answers against the existing process.
Check before use: Compare every consequential statement with the supplied source. Record corrections and require the responsible person to approve the result.
5. Community-reviewed interpretation
For Maya, Garifuna, Kriol or other community content, ask who has authority to approve the material and its use. Do not generate restricted knowledge or treat a language model as a competent cultural representative.
Starter prompt
Edit this authorised community text for clarity. Preserve names, meaning and attribution. Flag terminology for a competent local reviewer. Do not add traditions, history, translations or claims of community endorsement.
Check before use: Compare every consequential statement with the supplied source. Record corrections and require the responsible person to approve the result.
Evaluate a small pilot
Record the existing task time, errors and costs. Test on a limited set of representative examples and count review/correction time as part of the work. Include cases where information is missing or contradictory. Stop if the system exposes private data, invents evidence or makes an unauthorised commitment.
Report the task, sample, dates, tool version, reviewer and observed results. Do not turn a drafting-time improvement into a national GDP, revenue, yield or health-impact claim. Set a budget from actual quotes rather than an unsupported regional estimate.
Sources and local checks
The current Financial Services Commission website and laws register are starting points for verifying financial-services requirements. This guide does not retain the former IFSC/IBC legal checklist as current advice. For export work, use product-specific sources and the export preparation guide.
For other tasks, maintain a small source register with the issuing body, exact document, publication/effective date, date checked and unresolved questions. Verify local institutions and programme availability before including them. Ask competent local speakers to review translations. Keep an offline process for essential work if the selected tool or connection fails.
Related guides
Attribution
Cite as: Dunkley, A. (2026). Practical AI work in Belize. LAC AI Playbook. MaestrosAI.