Prompt Engineering Guide — Write Better AI Prompts

What is Prompt Engineering?

Prompt engineering is the practice of designing and optimizing the text inputs (prompts) you give to AI language models to get the best possible outputs. It's the interface between human intent and AI capability — the better your prompt, the better the AI's response.

Think of it like giving instructions to a very capable but literal-minded assistant. If you say "write something about dogs," you might get anything from a children's story to a veterinary manual. But if you say "write a 500-word blog post about why golden retrievers make great family pets, targeting first-time dog owners, in a friendly and informative tone," you'll get exactly what you need.

Prompt engineering isn't about tricking the AI or finding secret keywords. It's about clear communication — providing the right context, constraints, and examples to guide the model toward your desired output.

Why Prompt Engineering Matters

The difference between a good prompt and a bad prompt can be dramatic. Here's why it matters:

• Quality: A well-crafted prompt can produce output that needs minimal editing, while a vague prompt often requires extensive rework
• Consistency: Good prompts produce consistent results across multiple attempts
• Efficiency: Getting the right output on the first try saves time and API costs
• Capability Unlocking: Models can do things that aren't obvious without the right prompting technique

Studies show that prompt engineering can improve output quality by 30-50% for complex tasks. For professional use cases, this translates directly to productivity gains.

Core Principles

Before diving into specific techniques, understand these foundational principles.

1. Be Specific

Vague prompts produce vague outputs. Always specify:

• What: The exact task you want performed
• Format: How you want the output structured (paragraph, list, table, code)
• Length: Approximate word count or number of items
• Audience: Who the output is for
• Tone: Formal, casual, technical, friendly

2. Provide Context

Give the model the background information it needs. This includes:

• Your role or situation
• What you're trying to achieve
• Any constraints or requirements
• Relevant background information

3. Use Structure

Well-structured prompts are easier for models to follow:

• Use numbered lists for multi-part requests
• Separate instructions from context with clear delimiters
• Use headers or labels for different sections

4. Iterate and Refine

Your first prompt rarely produces the perfect output. Treat prompting as an iterative process — analyze the output, identify what's missing or wrong, and refine your prompt accordingly.

Role Prompting

Role prompting is one of the most powerful and easy-to-use techniques. You assign the AI a specific role or persona, which influences its knowledge, tone, and approach.

How It Works

Start your prompt with a role definition:

"You are a senior software engineer with 15 years of experience in Python. You write clean, well-documented code and follow PEP 8 conventions."

The model will then respond in character, drawing on the knowledge and style associated with that role.

Examples

• Technical writing: "You are a technical documentation writer at a major tech company. Write clear, concise documentation that junior developers can understand."
• Code review: "You are a senior code reviewer. Be thorough but constructive. Focus on potential bugs, performance issues, and code style."
• Creative writing: "You are a bestselling thriller novelist. Write with tension, vivid descriptions, and surprising plot twists."
• Business analysis: "You are a management consultant at McKinsey. Analyze this business problem using structured frameworks."

Tips for Effective Role Prompting

• Be specific about the role's expertise and experience level
• Include the role's communication style and priorities
• Combine with other techniques (e.g., "As a senior engineer, use chain-of-thought reasoning to...")

Chain of Thought

Chain of Thought (CoT) prompting asks the model to show its reasoning process step by step. This dramatically improves performance on tasks that require logic, math, or complex analysis.

Basic CoT

Simply add "Think step by step" or "Show your reasoning" to your prompt:

"A store has 15 apples. They sell 7 in the morning and receive a shipment of 12 in the afternoon. How many apples do they have now? Think step by step."

Structured CoT

For complex problems, provide a reasoning framework:

"Analyze this business proposal. First, identify the key assumptions. Then, evaluate each assumption's validity. Next, assess the risks. Finally, provide your recommendation with supporting reasoning."

When to Use CoT

• Mathematical or logical reasoning
• Complex analysis tasks
• Multi-step problem solving
• Decision-making with multiple factors
• Debugging code

Read our full Chain of Thought guide →

Few-Shot Prompting

Few-shot prompting provides examples of the desired input-output format before asking the model to perform the task. This is one of the most effective techniques for getting consistent, high-quality outputs.

How It Works

Include 2-5 examples in your prompt:

"Classify the sentiment of these reviews:
'This product is amazing!' → Positive
'Terrible quality, broke after one day.' → Negative
'It's okay, nothing special.' → Neutral
'Best purchase I've ever made!' → ?"

Zero-Shot vs Few-Shot

Tips for Effective Few-Shot

• Use diverse examples that cover different cases
• Keep examples concise — don't waste context window
• Ensure examples are consistent in format
• Include edge cases if relevant

Prompt Templates

Here are proven templates for common use cases. Copy and customize them for your needs.

General Task Template

"[Role]: You are a [specific role].
[Task]: [Clear description of what you need]
[Context]: [Relevant background information]
[Format]: [How you want the output structured]
[Constraints]: [Any limitations or requirements]
[Examples]: [Optional examples of desired output]"

Code Generation Template

"You are an expert [language] developer.
Write a [function/class/script] that [specific requirement].
Requirements:
- [Requirement 1]
- [Requirement 2]
- [Requirement 3]
Include comments explaining the code. Follow [style guide] conventions."

Analysis Template

"You are a [domain] analyst.
Analyze the following [data/text/situation]:
[Input]
Provide:
1. Key findings
2. Underlying patterns
3. Implications
4. Recommendations
Support each point with evidence from the input."

Writing Template

"You are a [type] writer.
Write a [format] about [topic].
Target audience: [audience]
Tone: [tone]
Length: [word count]
Key points to cover:
- [Point 1]
- [Point 2]
- [Point 3]"

Advanced Techniques

Once you've mastered the basics, these advanced techniques can further improve your results.

Self-Consistency

Ask the model to generate multiple solutions and then evaluate which is best. This is useful for complex reasoning tasks:

"Generate three different approaches to solving this problem. For each approach, explain the reasoning. Then, evaluate which approach is best and why."

Tree of Thought

For complex problems, ask the model to explore multiple reasoning paths:

"Consider this problem from three different angles. For each angle, think through the implications step by step. Then synthesize the insights from all three perspectives into a final answer."

Meta-Prompting

Use the AI to improve your prompts:

"I'm trying to get AI to [goal]. My current prompt is: [prompt]. How can I improve this prompt to get better results?"

XML Tags for Structure

Claude especially responds well to XML tags for structuring complex prompts:

"<context>Background information</context>
<task>What you need done</task>
<format>Desired output format</format>
<examples>Example outputs</examples>"

Common Mistakes

Avoid these common pitfalls that lead to poor AI outputs.

1. Being Too Vague

Bad: "Write about dogs."
Good: "Write a 300-word blog post about the top 5 dog breeds for apartment living, targeting young professionals, in a friendly tone."

2. Overloading a Single Prompt

Bad: "Write a blog post, create a social media caption, design an email subject line, and suggest hashtags."
Good: Break complex tasks into separate prompts, each focused on one deliverable.

3. Ignoring the Model's Strengths

Don't ask a language model to do math without providing a calculator tool. Don't ask for real-time information without web access. Work with the model's capabilities, not against them.

4. Not Providing Examples

If you want a specific format or style, show the model what you mean. One example is worth a paragraph of description.

5. Giving Up After One Try

If the first output isn't right, refine your prompt rather than starting over. Small adjustments often yield big improvements.