Week 2 of 12 · 12 hours · Month 1

Writing good requests: role, context, task, format and examples

Objectives of the week
  • Write a complete request in five elements: role, context, task, constraints and expected format.
  • Improve an answer through successive iterations rather than starting over.
  • Build a library of reusable standard requests for your job and your team.

Keep the course at home: free PDF handout with exercises, key takeaways and quiz answers.

Download this lesson as PDF Complete manual (PDF)

1Why wording matters so much

An AI assistant knows neither your organisation, nor your programme, nor your donor, nor your audience. A vague request ("Write a report on our project") produces generic text full of empty phrases. A precise request produces a usable first draft.

People often talk about prompt engineering. For everyday professional use, it is not about magic formulas but about good briefing practice, the same as briefing a competent new colleague who is discovering the file.

Simple test: if you gave your request as it stands to a consultant who knows nothing about your project, could they do a good job? If not, neither can the AI.

2The five elements of a good request

  • Role: the expected perspective. "You are an experienced MEAL specialist in nutrition programmes." The role steers vocabulary and the level of rigour.
  • Context: who you are, the programme, the country, the audience, the constraints. "National NGO, nutrition programme funded by a European donor, three districts, interim report at 6 months."
  • Task: a clear action verb. "Write the 'Progress by result' section from the data below."
  • Constraints: length, tone, elements to include or exclude, rules. "400 words maximum, factual tone, do not invent any figure, flag missing data."
  • Format: table, list, paragraphs, headings, language. "A table indicator / target / achieved / comment, then two paragraphs."

Complete example. "You are the grants manager of a medical NGO. Context: we are responding to a EUR 500,000 call for proposals for primary health care in two provinces; the donor requires an approach centred on accountability to affected populations. Task: propose a detailed outline of the 'Accountability' section (headings and subheadings, 3 points per subheading). Constraints: one page maximum, no promise we cannot keep, Core Humanitarian Standard terms. Format: numbered list."

3Giving examples and documents

The most reliable way to get the desired style is to show an example. A paragraph from a past report the donor liked, an email template, a well-completed indicator sheet: the assistant will imitate the structure and tone.

For long reference documents, use clear markers: "Here is the logframe between the tags <logframe> and </logframe>, and the quarter's data between <data> and </data>." Both assistants handle this kind of structuring well.

Explicitly ask the assistant to rely only on the documents provided when accuracy matters: "If information is not in the documents, write 'information not available' instead of guessing it." This sentence clearly reduces invention.

Caution. Before pasting a document, check that it contains no personal or sensitive data (week 3). A past donor report sometimes contains beneficiary names, partner lists or security incidents.

4Iterating: the conversation is a workshop

The first answer is rarely the right one. Rather than rewriting everything, correct step by step:

  • "Shorten to 250 words, keeping the three key figures."
  • "The tone is too promotional; make it factual, like a monitoring report."
  • "The second paragraph repeats the first: merge them."
  • "Add a line on delays due to the rainy season, with their effect on indicator 2.1."

Two very effective techniques:

  • Ask for questions first: "Before drafting, ask me the 5 questions you need to do a good job." The assistant thus reveals what your brief is missing.
  • Ask for a critique: "Review this text as a demanding donor evaluator: list the 5 main weaknesses, from most to least serious." Then correct it yourself or with the AI.

When a conversation drifts or becomes very long, start a fresh conversation with an improved brief: it is often faster.

5Standing instructions: Projects and Gems

If you often repeat the same context, do not paste it every time:

  • in Claude, a Project brings together standing instructions and reference documents (for example the communication charter, the logframe, the glossary) shared by all conversations in the project; depending on the plan, a project can be shared with colleagues;
  • in Gemini, a Gem is a custom assistant with its instructions and, depending on the version, reference files.

Example of standing instructions for a communications officer: "You help the communications team of NGO X. Tone: sober, respectful, no pity-based imagery. Never identify a child or a victim. Always use 'internally displaced people' and not 'refugees' for IDPs. Default length: 150 words. Flag any statement that requires a source."

These instructions embody the organisation's rules. They must be reviewed by the responsible person, like any procedure document.

6A request library for the team

The teams that get the most value from AI share their requests that work. A simple library (a shared document or spreadsheet) contains for each standard request:

  • the name and use ("Summary of the minutes of a cluster meeting");
  • the text of the request, with fields to fill in between square brackets;
  • the documents to provide and those never to provide;
  • the mandatory checkpoints ("check dates and organisation names");
  • the author and date of last update.

Start small: 10 requests for the most frequent tasks of the job (emails, minutes, summaries, rephrasing, translations). Measure the time saved and the quality, then expand.

A good request is worth a good document template: it can be shared, tested, improved and filed with the procedures.

Exercise: turning a vague request into a professional brief

Situation. A logistics coordinator writes to the assistant: "Write me an email for the supplier, he's late." The result is polite but useless: no order number, no date, no consequence.

Task. Rewrite the request with the five elements (role, context, task, constraints, format), inventing plausible data.

Possible answer. "You are an experienced logistics manager in an NGO. Context: purchase order PO-2026-114 of 2 October covers 400 hygiene kits; delivery was due on 30 October at the latest under the contract; today is 12 November; distribution to displaced families is scheduled for 20 November; the contract provides for penalties of 0.5% per day of delay, capped at 10%. Task: write an email to the supplier. Constraints: firm and courteous tone; request a written delivery date within 48 hours; recall the penalty clause without threatening; mention that we may cancel the order if delivery has not taken place before 17 November; 150 words maximum. Format: email subject then body, in English." Human check: order number, dates, penalty rate and cancellation clause consistent with the signed contract.

Key points

  • Brief the AI like a competent new colleague who knows nothing about the file.
  • A good request contains a role, a context, a task, constraints and a format.
  • An example and documents provided give the best result; require 'information not available' rather than an invention.
  • Iterate: ask for questions first, then a critique, then targeted corrections.
  • Projects (Claude) and Gems (Gemini) keep standing instructions; share a library of tested requests.
A question about this module?

Ask the other participants: they are taking the same course as you.

Ask my question See this module's discussions

💼 Programme management jobs

Put this course into practice: the latest jobs in this field.

See all 1079 Programme management jobs →

Prepare for the written test and interviewA model written test in this field, interview questions with sample answers and the STAR method.See the model test →
Further reading: course references
  1. ICRC and Brussels Privacy Hub (2020). Handbook on Data Protection in Humanitarian Action (2nd edition). www.icrc.org
  2. Inter-Agency Standing Committee (IASC) (2023). Operational Guidance on Data Responsibility in Humanitarian Action. interagencystandingcommittee.org
  3. UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence. www.unesco.org
  4. United Nations (Chief Executives Board) (2022). Principles for the Ethical Use of Artificial Intelligence in the United Nations System. unsceb.org
  5. European Union (2024). Regulation (EU) 2024/1689 on artificial intelligence (AI Act). eur-lex.europa.eu
  6. European Union (2016). General Data Protection Regulation (GDPR). eur-lex.europa.eu
  7. CHS Alliance, Groupe URD and Sphere (2024). Core Humanitarian Standard on Quality and Accountability. www.corehumanitarianstandard.org
  8. Sphere Association (2018). The Sphere Handbook: Humanitarian Charter and Minimum Standards in Humanitarian Response. spherestandards.org
  9. OECD (2019). Better Criteria for Better Evaluation: revised DAC evaluation criteria and principles for use. www.oecd.org
  10. National Institute of Standards and Technology (NIST) (2023). AI Risk Management Framework (AI RMF 1.0). www.nist.gov
  11. Anthropic (2026). Claude documentation and help centre (features, privacy, plans). support.anthropic.com
  12. Google (2026). Help centre for Gemini Apps and Gemini for Google Workspace. support.google.com/gemini
  13. Anthropic (2026). Claude prompting guide (developer documentation). docs.anthropic.com
  14. OCHA (2026). ReliefWeb: humanitarian reports and maps. reliefweb.int

Weekly quiz

5 demanding questions to check your learning. Answer them all to complete the week.

1Which is NOT one of the five elements of a good request?
2To reduce inventions when accuracy matters, you ask:
3"Before drafting, ask me the questions you need" is used to:
4In Claude, the space that brings together standing instructions and reference documents is called:
5A team request library should contain in particular:

← Week 1Week 3 →