Introduction: Why Prompts Matter

These days, it's harder to find someone who hasn't used AI like ChatGPT or Claude than someone who has. But when you actually use them, you often find yourself thinking, "Why isn't it giving me what I want?" Some people get amazing results with the same AI, while others only get irrelevant responses. That difference lies in the 'prompt'.

When I first encountered ChatGPT, I thought I could just ask casually. I assumed if I said "write a blog post," it would figure everything out. But reality was different. The output was either too generic or went in a completely different direction than I intended. So I began seriously studying how to write prompts, and I want to share what I've learned through this series.

What is Prompt Engineering?

Prompt Engineering is simply the systematic design of how you question or instruct an AI. In English, 'prompt' means 'to urge' or 'to encourage,' but in the AI context, it refers to the entire text input that a user provides to the AI.

Why such a grand term as 'engineering'? Because it's different from simply being good at communication. Like designing a machine, it's about designing inputs to get optimal results from the AI. A good prompt can bring out 120% of an AI's capability, while a bad prompt can make even the latest AI useless.

Let me give you an interesting analogy. Think of AI as a very smart new employee who can't read context. If you tell this new hire "handle that thing we talked about last time," they obviously won't understand. But if you say "create a draft proposal for Company A that we discussed in yesterday's meeting, 5 pages by next Monday. Key points are cost reduction and delivery time reduction," they can do the job accurately.

5 Principles of Good Prompts

Here are the 5 core principles of good prompts that I've compiled through countless trial and error. Following these principles alone will dramatically improve your results.

1. Clarity

This is the most basic yet most overlooked principle. AI cannot read your mind. Ambiguous expressions lead to ambiguous results.

Bad example: "Write a good article"

Good example: "Write a blog post about morning routines for office workers. Focus on the benefits of waking up at 5 AM and include 3 practical tips."

To increase clarity, ask yourself: "If someone else reads this prompt, would they understand it the same way?" If there's room for interpretation, you need to be more specific.

2. Specificity

This seems similar to clarity but is slightly different. Specificity is about providing details regarding 'what, how much, and in what format.'

Bad example: "Write a marketing email"

Good example: "Write a welcome email for new customers. About 200 characters, friendly tone, emphasizing 10% discount on first purchase. Also suggest 3 CTA button phrases."

Use numbers actively. "About 200 characters" is much more effective than "keep it short," and "3 items" is better than "a few."

3. Context

AI doesn't know who you are or why you need this. Providing background information leads to much more appropriate results.

Bad example: "Give me feedback on my resume"

Good example: "I'm a frontend developer with 5 years of experience. I'm preparing to move to a major company like Google or Amazon. Please review the resume below and tell me improvements from a big tech recruiter's perspective. [Resume content]"

Context can include your situation, goals, target audience, purpose of use, etc. The more relevant background information you provide, the better.

4. Examples

What's hard to explain in words can be understood instantly with a single example. The same goes for AI. If you show examples of the desired output, AI learns that pattern and produces similar output.

Bad example: "Write short phrases for social media"

Good example: "Write short copy for Instagram. Here's the style: 'A moment of peace with a cup of coffee today, small happiness adds up to great strength.' Create 3 phrases like this about 'reading'."

Examples are especially effective when you want to match tone, format, and length.

5. Constraints

Specifying what AI should NOT do and what limits to observe is also important. Without this, AI may interpret too freely and go in unwanted directions.

Bad example: "Tell me health information"

Good example: "Tell me dietary tips for diabetes prevention. However, only provide general health information, not medical advice. Don't mention medications or treatments. Organize 5 easy-to-implement daily habits."

Constraints are particularly useful for explicitly excluding unwanted things. Use expressions like "don't do~", "except for~", "only~".

3 Basic Prompt Patterns

Now let me introduce 3 prompt patterns you can use immediately in practice. These patterns are even more effective when combined.

Pattern 1: Role Assignment

This method assigns a specific role or persona to the AI. This way, it responds with expertise and perspective appropriate to that role.

You are a senior copywriter with 10 years of experience.
Answer questions about advertising copy from an expert perspective.
Suggest 5 slogans for a skincare brand targeting Gen Z.

Role assignment is especially effective when you need professional advice, analysis from a specific perspective, or creative work. Use expressions like "You are~", "From the perspective of~", "Act like~".

Pattern 2: Step-by-Step Instructions

This method breaks down complex tasks into multiple steps. It guides AI to think carefully as it progresses.

Please create a business plan draft following these steps:

Step 1: First, briefly analyze the current pet supplies market
Step 2: Summarize our business item's (eco-friendly pet food) competitiveness in 3 points
Step 3: Identify target customers and their characteristics
Step 4: Based on the above, create a business plan outline

This pattern shines especially in complex analysis or creative work. It prevents AI from trying to process too much at once and losing quality, and you can adjust direction by checking results at each step.

Pattern 3: Output Format Specification

This method specifies the format of the output in advance. If you specify the desired format such as tables, lists, JSON, or markdown, it outputs accordingly.

Please compare and analyze these 3 smartphones: iPhone 15, Galaxy S24, Pixel 8

Output in this format:
| Item | iPhone 15 | Galaxy S24 | Pixel 8 |
|------|-----------|------------|---------|
| Price | | | |
| Camera | | | |
| Battery | | | |
| Pros | | | |
| Cons | | | |

At the end, summarize in one sentence which type of user each product is suitable for.

Output format specification is particularly useful when you need to process data later or use it directly elsewhere. It can save significant time in situations requiring specific formats like reports, presentations, or code.

Common Beginner Mistakes and Solutions

When I first started learning prompt writing, I made countless mistakes. Based on that experience, I've compiled the most common mistakes and their solutions.

Mistake 1: Too Short Prompts

One-liners like "write an article," "write code," "translate this" are typical examples. AI is not a mind reader. With insufficient information, it can only produce the most generic, safest response.

Solution: Think of the 5W1H. Include as much as possible: Who (for whom), What, When (deadline), Where (in what context), Why (purpose), How (format).

Mistake 2: Asking for Too Much at Once

Requesting multiple things in one prompt like "write a blog post, optimize SEO, give thumbnail ideas, and create social media promo copy" degrades the quality of each.

Solution: Break up tasks. Have only one main goal per prompt. For related tasks, it's much better to request sequentially while continuing the conversation.

Mistake 3: Giving Up Without Feedback

Many people immediately write a new prompt or give up entirely when the first result isn't satisfactory. But since conversation with AI builds context, it's much more efficient to refine through feedback.

Solution: Develop results by giving feedback like "rewrite in a friendlier tone," "the second paragraph is too long, cut it in half," "add more examples."

Mistake 4: Using Only Negative Instructions

Negative statements like "don't make it boring," "don't make it too long" are hard for AI to interpret. It understands what NOT to do, but what it SHOULD do is unclear.

Solution: Express what you want in positive statements rather than negative ones. Instead of "don't make it boring," say "write humorously." Instead of "don't make it too long," say "write about 500 characters."

Mistake 5: Ignoring AI's Limitations

Asking AI for real-time information, leaving decisions requiring personal judgment, or expecting 100% accurate fact-checking is unreasonable.

Solution: Understand AI's strengths and weaknesses and use it for appropriate tasks. AI shows excellent performance in creative writing, idea brainstorming, drafting, and organizing information. On the other hand, tasks involving latest news, medical/legal advice, or fact-checking must always be verified by humans.

Learning Through Practical Examples

Theory alone isn't enough. Let's see through examples how improving prompts changes results.

Example 1: Email Writing

Before (Bad prompt):

Write an email to send to a business partner

After (Good prompt):

Write a business email for this situation:

Situation: Our company (ABC Electronics) will delay delivery by 3 days
Recipient: Mr. Kim Chulsoo, Manager at XYZ Distribution (major client)
Purpose: Ask for understanding and maintain trust relationship
Tone: Polite but not stiff
Include: Apology, reason for delay (parts supply issue), new delivery date, promise to prevent recurrence
Length: 200-250 characters

Example 2: Code Writing Request

Before (Bad prompt):

Create a login feature

After (Good prompt):

Implement a login feature in Python Flask with these requirements:

Tech stack: Python 3.11, Flask, SQLAlchemy, bcrypt
Feature requirements:
- Email/password based login
- Encrypted password storage
- Return error message on login failure
- Session-based authentication

Add comments to explain what each part does.
Also leave comments about security considerations.

Example 3: Blog Post Planning

Before (Bad prompt):

Recommend blog topics

After (Good prompt):

Recommend 5 blog post topics that meet these conditions:

Blog type: Self-improvement blog for 30-something office workers
Monthly visitors: About 5,000
Popular previous posts: "How to Use 1 Hour After Work," "Burnout Recovery Story"
Topics to avoid: Investing, dieting (already covered extensively)
Goal: Practical topics with high search traffic + shareability

For each topic:
- Title suggestion
- Expected target keywords
- 3 key points for the article
Please organize this.

Example 4: Data Analysis Request

Before (Bad prompt):

Analyze this data: [data]

After (Good prompt):

Analyze the sales data below:

[Paste data]

Analysis purpose: Planning next quarter's inventory
What I want to know:
1. Monthly sales trends and seasonality patterns
2. Top 5 best-selling products
3. Products with low inventory turnover relative to sales
4. Expected demand for next quarter (rough estimate level)

I will present the results at the marketing team's weekly meeting,
so explain insights in an easy-to-understand way for non-experts,
and organize key insights in bullet points.

Conclusion: Practice is the Answer to Prompt Engineering

Prompt engineering is ultimately about 'practice.' Don't just keep today's principles and patterns in your head - apply them right away when you request something from AI. It may seem to take more time at first, but once you get used to it, you'll get the results you want much faster and more accurately.

One more tip: save good prompts. I collect frequently used prompts in Notion or a notes app and pull them out when needed. Over time, you'll build your own prompt library, and this becomes an incredible productivity asset.

In Part 2, we'll cover prompt writing techniques and advanced usage strategies specifically for ChatGPT. We're preparing more in-depth content including Custom Instructions setup, GPTs usage, and practical examples for different work tasks, so stay tuned.

Series Guide

This article is Part 1 of the 'AI Prompt Engineering Practical' series.

  • Part 1: Prompt Engineering Basics - How to Communicate Effectively with AI (Current article)
  • Part 2: Mastering ChatGPT - Practical Usage Strategies