Prompt Engineering 101

The difference between a good AI response and a bad one is often just how you ask.

Here’s how to ask better.

The Core Principle

Vague input = vague output

Bad: "Write about dogs"
Good: "Write a 300-word blog post about why golden retrievers make good family pets"

The AI doesn’t know what you want unless you tell it.

Five Essential Techniques

1. Be Specific

Bad:

Help me write an email

Good:

Write a professional email to a client explaining
that their project will be delayed by one week
due to supply chain issues. Offer a 10% discount
as compensation. Keep it under 150 words.
Tone: Apologetic but confident.

Include:

• What you want
• Context/background
• Constraints (length, tone, format)
• Who it’s for

2. Give Examples

Show what you want:

Rewrite these sentences to be more concise:

Example:
Input: "Due to the fact that it was raining outside, we decided to stay home"
Output: "We stayed home because of the rain"

Now rewrite these:
1. "In spite of the fact that he was tired..."
2. "At this point in time, we are..."

Examples clarify expectations better than explanations.

3. Assign a Role

You are a senior software engineer reviewing code.
Look at this function and identify potential bugs,
performance issues, and security concerns.

Roles help AI:

• Adopt appropriate expertise level
• Use correct terminology
• Apply relevant knowledge

4. Structure Your Request

Use clear formatting:

Task: Write a product description

Product: Wireless earbuds
Target audience: Fitness enthusiasts
Key features:
- 8 hour battery life
- Sweat resistant
- Noise cancellation

Requirements:
- 100-150 words
- Conversational tone
- Include call to action

Output format: Plain text paragraph

Structure prevents misunderstanding.

5. Iterate and Refine

First attempt rarely perfect. Follow up:

Good, but:
- Make the opening stronger
- Remove the cliché in paragraph 2
- Add more specific examples

Work with the AI, not against it.

Advanced Techniques

Chain of Thought

Make AI think step by step:

Solve this problem step by step:

A store has 150 apples. They sell 40% on Monday,
then receive 30 more on Tuesday. How many do they
have Wednesday morning?

Show your work.

Forces clearer reasoning, fewer errors.

Few-Shot Learning

Multiple examples establish patterns:

Convert these casual phrases to formal business language:

"Gonna need that ASAP" → "Please prioritize this for urgent delivery"
"That's a no-brainer" → "The decision is straightforward"
"Let's touch base" → "Let's schedule a meeting to discuss"

Now convert:
"We need to get on the same page"

Negative Prompting

Specify what NOT to do:

Explain machine learning to a beginner.

Do NOT:
- Use technical jargon without explaining it
- Write more than 3 paragraphs
- Assume prior knowledge of statistics

Boundaries prevent unwanted output.

System Prompts

Set persistent context:

System: You are a technical writer creating documentation.
You write clear, concise explanations with practical examples.
You avoid jargon unless necessary, and define terms when first used.

Available in ChatGPT (custom instructions) and API.

Format Specifications

For Structured Output

Respond in JSON format:
{
  "summary": "...",
  "key_points": ["...", "..."],
  "action_items": ["...", "..."],
  "sentiment": "positive/negative/neutral"
}

For Lists

Provide your answer as a numbered list with
brief explanations for each point (2-3 sentences).

For Tables

Present this comparison as a markdown table with columns:
Feature | Tool A | Tool B | Recommendation

For Specific Length

Keep response to:
- 3 short paragraphs
- Under 200 words
- Tweet length (under 280 characters)

By Task Type

For Writing

Write a [type] about [topic].

Audience: [who will read this]
Tone: [casual/professional/technical/friendly]
Length: [word count or page equivalent]
Goal: [inform/persuade/entertain/explain]

Include:
- [required element]
- [required element]

Avoid:
- [thing to avoid]

For Analysis

Analyze [subject/document/code/situation].

Focus on:
- [specific aspect]
- [specific aspect]

Provide:
- Key findings
- Potential issues
- Recommendations

Be direct. Don't hedge unnecessarily.

For Brainstorming

Generate [number] ideas for [purpose].

Context: [relevant background]
Constraints: [limitations]

For each idea:
- Brief description
- Pros
- Cons
- Effort estimate

Include some unconventional options.

For Problem-Solving

I'm trying to [goal] but [obstacle].

Background:
- [context]
- [what I've tried]
- [constraints]

Help me think through this problem.
What questions should I be asking?
What approaches might work?

Common Mistakes

Being Too Vague

Problem: “Make this better”

Fix: “Improve this by [specific criteria]. Focus on [specific aspect].”

Over-Explaining

Problem: 500 words of context for a simple question

Fix: Include only relevant information

Not Iterating

Problem: Accepting first output even when imperfect

Fix: Follow up with specific feedback

Wrong Format Request

Problem: Asking for essay when you need bullets

Fix: Specify output format explicitly

Assuming Knowledge

Problem: Using acronyms or internal jargon

Fix: Define terms, provide context

Prompts That Work

Content Creation

Write [content type] about [topic] for [audience].

Requirements:
- [length]
- [tone]
- [specific elements to include]

Make it [quality descriptor] and [quality descriptor].

Code Help

[Language]: [what you're trying to do]

Current code:
[paste code]

Problem:
[what's happening vs. what should happen]

Help me fix this and explain what was wrong.

Learning

Explain [concept] to me.

My background: [relevant experience level]
I learn best through: [examples/analogies/step-by-step]

Start simple, then go deeper.

Decision Making

Help me decide between [option A] and [option B].

Context: [situation]
My priorities: [what matters most]
Constraints: [limitations]

Give me a clear recommendation with reasoning.

Testing Your Prompts

Does it work?

Run it 3 times. Are results:

• Consistently good? → Prompt is solid
• Sometimes good? → Needs refinement
• Always bad? → Rethink approach

Quick Improvements

1. Add more specifics
2. Include examples
3. Specify format
4. Add constraints
5. Break into steps

The Bottom Line

Good prompting is about clear communication:

1. Know what you want before asking
2. Be specific about requirements
3. Show examples when possible
4. Specify format explicitly
5. Iterate on results

Small improvements in prompting yield dramatic improvements in output. Worth investing 2-3 sentences of additional context.

The AI is capable. Your job is to unlock that capability with clear instructions.