Learning goals
- You build prompts from six building blocks.
- You improve a prompt deliberately over several rounds.
- You set up a small prompt library for your team.
The prompt toolkit
Good prompts are not magic spells; they are good work briefs. They contain the same information you would give a new colleague. Six building blocks have proven useful:
| Building block | Question | Example |
|---|---|---|
| Role | From which perspective? | "You are a customer service pro at an online shop." |
| Goal | What should the result be? | "Write a reply to this complaint." |
| Context | What does the model need to know? | Customer email, return policy, order status |
| Format | What should it look like? | "Five sentences at most, informal, no small talk." |
| Examples | What should it follow? | A good past reply |
| Checks | How do you know it worked? | "State the deadline. Do not invent commitments." |
Not every prompt needs all six. Goal, context and format almost always belong in it.
Iterate instead of starting over
The first answer is rarely the best. Instead of discarding the prompt, tell the model exactly what is missing: "Shorter", "No jargon", "Add the costs as a table". Precise feedback works better than a fresh attempt.
Three techniques help most:
- Let it ask first: "Ask me three questions before you write." Missing context surfaces.
- Separate steps: Outline first, then text. Analysis first, then recommendation.
- Self-check: "Check your answer against the criteria and mark what is not met."
The prompt library
A good prompt is a work tool and should be saved like a template. Set up a simple library for your team, as a spreadsheet or wiki page. Five fields per entry are enough: task, prompt, context needed, permitted data class, contact person.
Many vendors maintain their own prompt collections, including Microsoft, Google, Anthropic and OpenAI. They are a good starting point but rarely fit your company without changes. The best entries come from your team's real tasks.
Three rounds for one prompt
Improve a prompt for a task you do every week.
Do the task in your day-to-day work
You do this task in your day-to-day work, not here in the course. Tick off the steps once you have done them.
What comes out of it: Your first entry in your team's prompt library.
Enter your result here
Enter what came out of the task. Everything is saved automatically.
See an example
This is what a filled-in result could look like. Your entries stay unchanged.
Your prompt in three rounds
- Task the prompt is for
- Draft replies to customer complaints
- Round 1: first prompt
- Write a reply to this complaint.
- What was missing?
- Too formal, no solution offered
- Round 2: with goal, context and format
- You are a customer service professional at an online shop. Goal: a friendly reply with a solution. Context: the delivery arrived damaged. Format: no more than 120 words, informal tone.
- What improved?
- Solution included, tone fits
- Round 3: best version
- You are a customer service professional at [company]. Reply to the complaint below. Goal: acknowledge the problem, offer a solution (replacement or credit), name the next steps. Format: no more than 120 words, informal tone, no filler. Ask if the order number is missing.
- Context needed
- Complaint text, order number, goodwill rules
- Permitted data class
- Internal
- Contact person
- Customer service team lead
Kept in your workbook
No later module needs this result. It stays in the workbook, where you can export it.
Quiz
1. Which three building blocks almost always belong in a business prompt?
2. The first answer is too long and too generic. What is the best next step?
3. What is the instruction "Ask me three questions before you write" for?
4. Which field belongs in every prompt library entry so it is used safely?