Prompt Engineering
How to ask an AI for what you need: the five parts of a clear request, and how to fix a mediocre answer instead of settling for it.
- 3 sections
- about 4 minutes
- updated October 2026
Written to be quoted, by people and by AI assistants. This guide is also plain text at /ai-101/prompt-engineering.md.
Why Prompts Matter
The quality of your prompt directly determines the quality of the AI’s response. A vague prompt gets a generic answer. A clear prompt gets a useful one. Prompt engineering is the skill of communicating your intent clearly to an AI system.
Same model, different answer
Ask one model the same question twice, once in a line and once with the job, the reader and the limits spelled out, and you get two different answers. The model did not change. What you told it did.
Think of prompting as briefing a new hire on day one. They do what you say and know nothing about your business, so say what you want, who it is for, and what a good result looks like.
Anatomy of a Great Prompt
Every effective prompt contains some combination of these five elements. You don’t always need all of them, but the more complex your task, the more elements you should include.
- Role: Who should the AI act as? (e.g., “You write customer emails for a family-run plumbing company in Baton Rouge…”)
- Task: What specific thing should the AI do? (e.g., “Draft a reply to this customer asking when we can come out…”)
- Context: What background information is relevant? (e.g., “We are booked through Thursday, and a burst pipe goes ahead of the queue…”)
- Format: How should the response be structured? (e.g., “Two short paragraphs, then the open appointment times as a list…”)
- Constraints: What limitations or boundaries apply? (e.g., “Under 120 words. Do not quote a price…”)
# Weak prompt:
"Write a follow-up email"
# Strong prompt:
"You write client emails for a plumbing company. Draft a
follow-up to a client who received our quote ten days ago
and has gone quiet.
- Mention the quote date and the job (bathroom refit)
- Friendly, no pressure, one question at the end
- Under 120 words, no exclamation marks"Check yourself
Which element of a prompt helps prevent overly long or off-topic responses?
Iterative Refinement and Evaluation
A first prompt is a draft. Write it, run it, read what comes back, and fix the part that missed. Most of the work happens in that loop.
- Start simple: Begin with a basic prompt that captures your core intent
- Test it: Run the prompt and carefully read the output
- Identify gaps: What’s missing? What’s wrong? What’s unnecessary?
- Add specificity: Address each gap with more precise instructions
- Test variations: Try different phrasings, structures, and frameworks
- Evaluate consistently: Use the same test cases to compare prompt versions
Where the effort goes
A basic prompt often gets you most of the way. Getting the output exactly right is where examples, a fixed format, and clear constraints do the work.
Check yourself
What’s the most effective way to improve a prompt that’s giving mediocre results?
Responsible AI Use