What is prompt engineering?
A prompt is the instruction you give an AI model. Prompt engineering is the practice of crafting those instructions to get better, more reliable outputs.
The good news: most effective techniques are simple once you know them.
1. Be specific and explicit
Vague prompts produce vague output.
Bad: Write about climate change.
Better: Write a 200-word plain-English explainer on how CO₂ causes the greenhouse effect, aimed at a secondary school student with no science background.
The better version specifies: format (200 words), audience (secondary school student), topic (CO₂ and the greenhouse effect), and tone (plain English).
2. Give the model a role
Assigning a role primes the model's "perspective".
You are an expert Python developer reviewing a junior developer's code.
Review the following function and explain any issues to the junior developer
in encouraging, constructive language.
3. Few-shot examples
Show the model what good output looks like with 2–3 examples before your actual request.
Convert these product descriptions to bullet points:
Input: "A sturdy, waterproof backpack with 30L capacity and laptop sleeve."
Output:
- Waterproof construction
- 30L capacity
- Dedicated laptop sleeve
Input: "Noise-cancelling headphones with 40-hour battery and USB-C charging."
Output:
- Active noise cancellation
- 40-hour battery life
- USB-C charging
Input: "Ergonomic office chair with lumbar support and adjustable armrests."
Output:
4. Chain-of-thought (CoT)
For reasoning tasks, ask the model to think step by step before giving a final answer.
A store sells apples for $0.50 each and bags of 6 for $2.50.
I want to buy 15 apples. What's the cheapest way?
Think through this step by step before giving your final answer.
This dramatically improves accuracy on maths, logic, and multi-step problems.
5. Specify the output format
Tell the model exactly how you want the response structured.
List three pros and three cons of remote work.
Format your response as:
Pros:
1. ...
2. ...
3. ...
Cons:
1. ...
2. ...
3. ...
What to try next
- Experiment with different temperatures (if the API lets you) — lower = more predictable, higher = more creative
- Try system prompts vs user prompts in the Anthropic docs
- Read our follow-up: Advanced Prompt Patterns (coming soon)