# Five Stages of Prompting

The teaching sequence is **Role → Task → Context → Constraints → Format**. Give the AI the details a new coworker would need before starting. Each stage removes a guess.

## Opening exercise: a new hire and a meeting room

Ask a new hire to get a meeting room ready for an afternoon meeting. They must ask which room, what time, how many people, and what equipment is needed. The unclear instruction caused the questions. Use the same observation when writing a request for AI: supply the facts and boundaries it cannot know on its own.

## 1. Role: who should answer?

Name the useful working perspective, audience, and expertise level. This shapes vocabulary, depth, tone, priorities, and scope. A role can stay useful across several related requests. It does not grant real credentials or replace a human expert.

## 2. Task: what should it do?

Use a specific verb such as summarize, compare, draft, or calculate. State what finished work looks like. Prefer one clear action to several vague requests stacked together.

## 3. Context: what facts does it need?

Give the relevant audience, situation, source material, prior decisions, and resources available. Context supplies facts, rather than new instructions. Include the facts that would change the answer; leave out unnecessary or unapproved details.

## 4. Constraints: what are the boundaries?

State limits such as length, tone, excluded topics, privacy requirements, and what to do when a fact is missing. Add these before the first response instead of correcting avoidable mistakes afterward.

## 5. Format: what should the result look like?

Specify the output shape: paragraphs, bullets, a table, an email, or a script. Add headings, columns, numbering, or section lengths when they will help someone use the result.

## Worked example


```text
Role: Act as a careful meeting coordinator writing for a busy project team.
Task: Summarize the decisions and follow-up work from these notes.
Context: These are approved, de-identified notes from a planning call. The readers missed the call and need the next steps. Use only the notes supplied below: [paste permitted notes].
Constraints: Do not invent a decision, owner, date, or quote. Mark an action item UNASSIGNED when no owner is named. Leave out personal details.
Format: Use three short sections headed Decisions, Action Items, and Open Questions. Put action items in a table with Owner, Action, and Due Date columns.
```

## Build it one piece at a time

Use the new example above to demonstrate the sequence:

1. Start vague: “Summarize this meeting.” The AI must guess what matters and who will read it.
2. Add **Role**: a meeting coordinator. The intended perspective and vocabulary sharpen.
3. Sharpen **Task**: extract decisions and follow-up work. The outcome becomes specific.
4. Add **Context**: intended readers and permitted notes. The response can be grounded in real material.
5. Add **Constraints**: no invented details or personal information. The boundaries become explicit.
6. Add **Format**: three sections and an action table. The result becomes ready to use.

Review the output with the source notes before anyone acts on it. The lesson is that better input removes guesses; it does not guarantee the answer is correct.