Writing Effective Prompts
Get useful output from an AI tool by supplying the context, role, format, and constraints it cannot infer.
By the end of this lesson you can
- Identify the components of a well-formed prompt
- Supply context and constraints that shape output usefully
- Iterate on a prompt rather than accepting a first attempt
- Recognize tasks where a prompt will not help
Lesson Notes
Read through the key concepts before you try the challenge.
A vague prompt gets a generic answer
You draft patient communications at Lakeside Medical Associates.
You type 'write a letter about missed appointments' and get four bland paragraphs that could belong to any organization, at a reading level most of your patients would struggle with, in a tone that sounds faintly like a debt collector. The tool was not wrong; it had nothing to work with.
Your task: Learn to supply the context that turns a generic response into a usable draft.
A model cannot infer your situation. It does not know your organization, your audience, your tone, or what a good answer looks like to you. Everything it does not know, it fills in with the most statistically average option — which is exactly why unspecified prompts produce bland output.
| Component | Supplies | Example |
|---|---|---|
| Role | The perspective to write from | You are an administrator at a small family medicine practice |
| Task | What you want, specifically | Draft a letter to a patient who missed a follow-up |
| Context | The situation and audience | The patient has a chronic condition needing regular monitoring |
| Format | The shape of the output | Three short paragraphs, under 200 words |
| Constraints | Rules it must respect | Plain language at a sixth-grade reading level; warm, never accusatory; no medical advice |
| Examples | What good looks like | Match the tone of this previous letter: [paste] |
Improving a prompt in three passes
Turn 'write a letter about missed appointments' into a prompt that produces something usable.
- 1
Add the role and the audience.
'You are an administrator at a small family medicine practice writing to a patient' immediately narrows register and vocabulary. Without it the model averages across every organization that has ever written about missed appointments, including debt collectors.
- 2
Add the constraints that matter most to you.
'Sixth-grade reading level, warm and non-accusatory, no medical advice, under 200 words.' Each constraint eliminates a failure mode you would otherwise have to fix by hand. The reading level constraint alone usually transforms the output.
- 3
Add an example of the tone you want.
Pasting a previous letter you were happy with communicates tone far more precisely than adjectives can. Models match demonstrated patterns much better than they follow descriptions of them.
- 4
Iterate on what is still wrong, specifically.
'The second paragraph sounds like a warning — rewrite it to emphasize that we want to help' is actionable. 'Make it better' is not. Treat it as a conversation with a competent drafter who cannot see your reaction.
Result: A draft in the right register and length, needing edits rather than a rewrite.
Role, task, context, format, constraints. Everything you leave unstated is filled in with the most average option available.
An AI tool returns generic, unusable text for a workplace writing task. What is the most likely cause?
Challenge
Apply what you've learned in this lesson.
Measure the difference context makes.
- Give an AI tool a one-line prompt for a workplace writing task. Save the output.
- Rewrite the prompt with all six components from the table. Save that output too.
- Compare them and list every specific improvement. Note how much editing each would need before sending.
- Build a reusable prompt template for a task you do regularly, with bracketed placeholders for the parts that change.
Finished this lesson?
Progress is saved in this browser only. It is not a grade — official progress lives in Brightspace.