Introduction: Are You Using ChatGPT Properly?

In Part 1, we covered the basic principles and patterns of prompt engineering. In this part, we'll dive deep into practical usage strategies focusing on ChatGPT, the most widely used AI. Honestly, if you're only using ChatGPT as a 'search replacement' or for 'simple Q&A,' you're utilizing less than 10% of this tool's potential.

I was the same way at first. I thought, "ChatGPT? Can't you just ask whatever you want?" But after setting up Custom Instructions, creating my own GPTs, and systematically applying them to my work, my perspective completely changed. When used properly, you can really get personal assistant-level help.

ChatGPT's Characteristics and Strengths

Let me first organize what characteristics ChatGPT has compared to other AI tools. You need to understand the tool's characteristics to develop appropriate strategies.

Conversational Context Retention

ChatGPT's biggest advantage is that it remembers conversational context. Within a single conversation session, it remembers previous discussions and can continue the dialogue building on that context. Using this well allows you to progress through complex projects step by step.

For example, let's say you're writing a business plan. In your first message, you explain the business idea and background, in the second message you request market analysis, and in the third message you request financial planning... ChatGPT maintains the context of the business idea you explained initially and creates consistent content throughout.

Support for Various Output Formats

You can receive results not just as text, but in various formats including code, tables, markdown, and JSON. Especially when you ask for output as code blocks, it organizes it in an easy-to-copy format, and when you ask for tables, it creates clean markdown tables.

Multimodal Capabilities (GPT-4o)

The latest model, GPT-4o, can understand and analyze images. You can show a screenshot and ask "What's wrong with this UI?", upload a photo of handwritten notes and say "Organize this," or show a graph image and say "Interpret this data." This significantly increases work efficiency.

Plugin and Integration Features

ChatGPT Plus users can utilize various features including web browsing, code execution (Code Interpreter), and image generation (DALL-E). Code Interpreter is particularly powerful for data analysis tasks. Upload an Excel file and it will analyze and visualize your data.

GPT-4 vs GPT-4o Comparison

Let's learn about the two main models you can choose from in ChatGPT. Depending on the situation, the quality of results can vary based on which model you select.

Category GPT-4 GPT-4o
Response Speed Relatively slow More than 2x faster
Multimodal Text-focused Integrated image and voice support
Reasoning Ability Excellent Equal to or slightly better than GPT-4
Cost (API) Relatively high 50% cheaper
Context Length 8K/32K 128K
Recommended Use Complex reasoning, long-form writing General conversation, image analysis, real-time use

In my personal experience, GPT-4o is sufficient for most everyday tasks. The fast response speed is great because it doesn't interrupt your workflow. However, for very complex logical reasoning or writing tasks where subtle nuances are important, GPT-4 seems to produce slightly better results.

Writing Prompts Optimized for ChatGPT

On top of the basic principles learned in Part 1, let me add some ChatGPT-specific tips.

Using System Messages

When starting a conversation with a role setting like "You are~", that role is maintained throughout the entire conversation. Using this well means you don't need to explain context every time.

You are a UX designer with 10 years of experience and a usability expert.
From now on, for any app screenshots or wireframes I show you,
give feedback from a usability perspective.
Organize feedback in the format "Strengths / Areas for Improvement / Specific Suggestions".

Starting like this, you can simply upload screenshots later and it will automatically provide feedback in the specified format.

Designing Conversation Flow

Don't request everything at once for complex tasks - design the conversation flow. This maximizes ChatGPT's context retention feature.

[Step 1 - Context Setting]
I'm a marketing manager at a B2B SaaS startup.
Our product is an HR management solution for SMBs,
and our main target is HR managers at companies with 50-200 employees.
Please remember this context.

[Step 2 - Specific Request]
In this context, plan a LinkedIn content series.
Suggest 8 content topics total, 2 per week for 4 weeks.

[Step 3 - Development]
I like the second topic. Create an actual post draft with that one.

[Step 4 - Refinement]
Make it a more casual tone. Add some emojis appropriately.

Using "Let's think step by step"

Having ChatGPT think step by step can yield more accurate answers. This is especially effective for math problems or questions requiring logical reasoning.

Solve the following problem. But don't give the answer right away -
show your thinking process step by step:

"Company A's revenue increased 20% year-over-year,
and Company B's increased 15%.
If Company A's revenue last year was 80% of Company B's,
what's the ratio of the two companies' revenue this year?"

Providing Few-shot Examples

When you show ChatGPT a few examples of your desired output, it learns that pattern and produces similar quality results.

Write product descriptions in the following format.

[Example 1]
Product: Wireless earbuds
Description: Adding music to your daily life. Your own soundtrack begins when they gently rest in your ears.
Your commute becomes a concert, and the cafe becomes a recording studio.

[Example 2]
Product: Smartwatch
Description: A smart assistant on your wrist. Messages, health, and schedules at a glance.
Check everything with a brief look amidst your busy day.

[Request]
Product: Power bank
Write a product description in the style above.

Using Custom Instructions

If you're a ChatGPT Plus user, Custom Instructions is a feature you must set up. This feature lets you pre-configure "information about yourself" and "desired response style" that applies to all conversations by default.

Where to Find Custom Instructions

Click on the profile icon at the bottom left of the ChatGPT screen and select "Customize ChatGPT." There are two input fields:

  • What would you like ChatGPT to know about you? - Information about yourself
  • How would you like ChatGPT to respond? - Desired response style

Effective Custom Instructions Examples

Information about me (example):

I'm a startup marketer in my 30s.
- Primarily handle content marketing for B2B SaaS products
- Main work involves creating blogs, newsletters, and social media content
- Data analysis skills: Intermediate Excel, basic SQL
- Can read English but struggle with business English writing
- Always busy so I value efficiency

Desired response style (example):

- Respond in English (include original text if needed)
- Key points first, explanations later (inverted pyramid structure)
- Skip unnecessary greetings or disclaimers
- Include specific examples and action items
- Use tables or lists for better readability
- Medium length by default, unless otherwise specified
- Prioritize practical advice from a marketing perspective

Setting this up means you don't have to repeat "respond in English," "just give me the key points" every time. Especially if you use ChatGPT repeatedly for work, the time-saving effect is enormous.

Creating and Using GPTs

GPTs is a feature for creating customized versions of ChatGPT for specific purposes. Anyone can create them without complex coding, and you can share created GPTs with team members or publish them on the GPT Store.

When You Should Create GPTs

  • When you repeatedly request similar types of tasks
  • When you need answers based on specific documents or data
  • When you need to maintain a consistent output format
  • When your entire team needs to use AI in a consistent manner

GPTs Creation Tutorial

For example, let's create a "Blog Post Outline Generator."

Step 1: Open the GPTs Editor

Click "Explore GPTs" in the ChatGPT left menu, then click the "+ Create" button in the top right.

Step 2: Basic Settings

Name: Blog Outline Master
Description: Creates SEO-optimized blog post outlines.
Enter a topic and get title suggestions, H2/H3 structure, and key points for each section.

Step 3: Writing Instructions

You are an SEO and content marketing expert.
When users input a blog topic, provide an outline in this format:

1. Title suggestions (3 options, click-inducing while including keywords)
2. Meta description (around 160 characters)
3. Target keywords (1 main + 3 secondary)
4. H2/H3 structure (minimum 5 H2 sections)
5. Key points for each section (3 bullet points each)
6. Internal link suggestions (3 related topics)
7. CTA (Call-to-Action) suggestions

Always organize outlines neatly in table format.
Respond in English and use natural, SEO-friendly expressions.

Step 4: Upload Knowledge Base (Optional)

If you upload your company's style guide, collection of existing blog posts, or target keyword list as PDF or text files, the GPT will reference these to create more consistent results.

Useful GPTs Ideas

  • Meeting Notes Organizer: Paste meeting recording content and get summary, action items, and assignee organization
  • Code Reviewer: Paste code to get potential bugs, improvements, and best practice suggestions
  • Email Tone Adjuster: Convert written email tone to polite/casual/firm
  • Interview Prep Coach: Input target company and position to get expected questions and answer guides
  • Data Interpreter: Paste Excel data to get insights and visualization suggestions

ChatGPT Usage Examples by Task

Now let's look at specific examples of how to use ChatGPT in actual work situations.

Writing Tasks

Use Case: Corporate blog post draft

Write a corporate blog post draft with the following conditions:

Topic: Effective team communication in remote work environments
Target readers: Startup founders and team leaders
Tone: Professional yet friendly, sharing experiences
Length: Around 2000 words
Include:
- 3 challenges of remote work communication
- Real solutions and tool recommendations for each challenge
- Natural mention of our product (collaboration tool SaaS company)
- Action items readers can implement immediately

Structure:
1. Engaging introduction (question or relatable situation)
2. Body (problem-solution format)
3. Product mention (not feeling like an advertisement)
4. Conclusion (encouragement + CTA)

Coding Tasks

Use Case: Code refactoring request

Refactor the Python code below.

Current code:
[paste code]

Requirements:
1. Follow PEP 8 style guide
2. Split function into smaller units
3. Add type hints
4. Improve error handling
5. Add docstrings

After refactoring:
- Show before/after code comparison
- Explain "why" each change was made
- Suggest any additional improvements

Data Analysis Tasks

Use Case: Marketing data analysis (using Code Interpreter)

[After uploading Excel file]

This data is marketing campaign performance data from the past 6 months.

Perform the following analysis:
1. Calculate and compare ROAS (Return on Ad Spend) by channel
2. Visualize monthly trends (line graph)
3. Selection criteria and list of TOP 5 best-performing campaigns
4. Identify campaigns with low cost efficiency
5. Budget allocation suggestions for next month

I'll be presenting these analysis results at the marketing team weekly meeting.
Organize insights at a level that non-experts can understand.

Translation Tasks

Use Case: Business email translation

Translate the email below to English.

Original:
[email content]

Translation requirements:
- Maintain formal business tone
- Convert local expressions to appropriate English business conventions
- Convert dates/times to appropriate timezone if needed
- For ambiguous translations, mark as [original: ~] and provide alternatives

Additionally:
- Point out any expressions that would seem awkward to native speakers
- Suggest any sentences that could be made more professional

ChatGPT's Limitations and Precautions

To use ChatGPT effectively, you also need to clearly understand its limitations. Blind trust can actually cause problems.

Fact-Checking is Essential

ChatGPT sometimes generates plausible but incorrect information. Specific numbers, dates, and quotes in particular must be verified separately. I've used statistics that ChatGPT provided directly, only to find later that the source was unknown or the numbers were different.

Tip: For important facts, ask "What's the source of this information?" If ChatGPT can't clearly provide a source, manual verification is needed.

Limitations on Latest Information

ChatGPT's training data has a cutoff date. Latest news, real-time stock prices, recently launched product information, etc. may not be accurate. Enabling web browsing helps somewhat, but for information where real-time accuracy is important, separate searches are still needed.

Personal and Confidential Information Caution

Content you input into ChatGPT may be used for model training. Company secrets, customer personal information, and sensitive business data should not be input as a rule. For enterprise use, ChatGPT Enterprise or API usage is recommended.

Tip: Turning off "Chat History & Training" in settings prevents your conversations from being used for training. However, this isn't complete confidentiality protection, so caution with sensitive information is still needed.

Copyright Issues with Generated Content

Copyright issues for AI-generated content are still not legally clear. For content to be used commercially, rather than using AI output directly, using it as reference material and having humans modify/supplement is recommended.

Beware of Over-Reliance

Relying too heavily on ChatGPT can weaken your own thinking and creativity. AI is a tool, and final judgment and responsibility rest with humans - don't forget this. Especially for important decisions, use AI output as reference, but always add your own judgment.

Conclusion: Next Steps in ChatGPT Usage

Let me summarize what we covered in this article:

  • ChatGPT's strengths: Context retention, various output formats, multimodal, plugins
  • GPT-4 vs GPT-4o: Choose based on situation
  • Prompt optimization: Role setting, conversation flow design, step by step, few-shot
  • Custom Instructions: Set up information about yourself + desired response style
  • GPTs: Create custom AI for repetitive tasks
  • Work applications: Specific examples for writing, coding, analysis, translation
  • Precautions: Fact-checking, personal information, avoiding over-reliance

There's no single right answer for using ChatGPT. Use the methods introduced here as reference and find your own usage style. What's important is to continuously experiment, save effective prompts, and keep refining them.

AI tools are developing rapidly. What you learned today may need updates in a few months. But the basic principles - communicating clearly, providing context, and describing what you want specifically - will remain valid no matter what AI emerges.

Series Guide

This article is Part 2 of the 'AI Prompt Engineering Practical' series.

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