AI Literacy for Humanitarian Professionals
This quick reference is for safe everyday use of AI in reports, summaries, proposals, planning notes, and learning products. It is not a replacement for professional judgment, safeguarding, data protection, or source verification.
Golden rule: if information could put a person, community, staff member, partner, organization, or operation at risk, do not enter it into a public AI tool.
The 5-Step Safe AI Routine
- Remove sensitive information. Strip names, contact details, locations, case details, identifiers, and unpublished internal records.
- Define the task. Tell the tool the audience, context, purpose, output format, and limits.
- Ask for separation. Require facts, assumptions, risks, missing information, and suggested next steps in separate sections.
- Verify claims. Check data, dates, laws, sources, program facts, donor rules, and contextual claims before using the output.
- Keep human approval. A named person must own the final text, decision, and consequences.
Prompt Formula
| Element | What to write |
|---|---|
| Role | Act as a careful humanitarian program officer / MEAL reviewer / proposal editor. |
| Task | Summarize, structure, improve, compare, draft, or identify gaps. |
| Context | Describe the project type, audience, document purpose, and constraints without sensitive details. |
| Output | Ask for bullets, table, memo, checklist, email, logframe comments, or action plan. |
| Quality rule | Ask the tool to flag uncertainty, missing evidence, assumptions, and claims needing verification. |
Do Not Enter
- survivor, beneficiary, household, child, medical, protection, or legal case details
- names, phone numbers, exact locations, identity numbers, staff files, or partner disputes
- security-sensitive field information, procurement details, unpublished donor data, or passwords/API keys
- confidential grant decisions, internal investigations, or financial irregularity details
Output Review Checklist
- Does the output invent facts, sources, numbers, organizations, dates, or commitments?
- Does it hide uncertainty or sound more confident than the evidence allows?
- Does it ignore affected communities, local actors, gender, disability, protection, or power dynamics?
- Does it create legal, safeguarding, donor-compliance, or reputational risk?
- Can a human reviewer explain and defend the final version?