5 Core Prompt Formulas That Make ChatGPT Read Your Mind (Practical Templates for Beginners)

5 Core Prompt Formulas That Make ChatGPT Read Your Mind (Practical Templates for Beginners)

"I clearly wrote the question this way—so why does ChatGPT give an answer that's completely different from what I intended?"

Have you had this experience? You asked for a draft marketing plan and got only generic content back, or you requested copy to hand off to a designer and the tone and manner were completely off.

Large language models (LLMs) like ChatGPT are truly all-purpose tools. But this powerful tool is not "magic"—it's a tool. Even the best tool can't perform at its full potential without a user manual.

We often feel a vague fear or hold the preconceived notion that "prompt engineering" is difficult. Don't worry. Prompt engineering isn't complex coding knowledge; it's closer to a communication skill that assigns AI a logical thinking sequence and a role.

This article is not a list of abstract theories. It contains 5 practical formulas that non-developer roles—marketers, planners, designers, and others who want to bring AI into their work—can copy and paste today and walk away thinking, "Wow, this actually works."

1. The Basics of Assigning AI a "Role" and "Constraints": Setting Persona and Context

The biggest reason AI gives off-base answers is that it doesn't know from whose perspective it should respond. By default, AI pretends to be an "expert in everything," so the depth and tone of its answers become vague.

The first thing to do is assign AI a clear persona (role) and constraints (context). It's like instructing a new hire: "You are a copywriter with 20 years of experience. The target for this product is women in their 30s who are beginner housewives. Never use technical jargon."

💡 Practical Template (Role Assignment):

[Role Assignment] You are a B2B SaaS specialist copywriter with 20 years of experience. [Target Setting] Our main target is IT managers at SMEs (men in their 30s–40s). They are most sensitive to "reducing work hours" rather than technical advantages. [Constraints] The answer must list only 3 key advantages concisely, and the use of jargon is prohibited.

When you specify the role like this, AI suddenly switches to "copywriter mode," and the tone and depth of the answers improve dramatically.

2. Advanced Techniques to Force AI's Thinking Process: CoT and Few-Shot Learning

Once you have the basics down, you need to teach AI how it should think. This stage is what elevates ordinary users to expert level.

🧠 Applying CoT (Chain-of-Thought)

CoT is not just asking AI for the final answer; it is asking it to show the "thinking process" itself. It's like not just writing the answer on an exam, but writing out the entire solution process.

❌ Bad Prompt: "Analyze this data and come up with a marketing strategy." (→ Only vague conclusions come out) ✅ CoT-Applied Prompt: "When analyzing the following data, you must think through the following 3 steps, explain that process in detail, and then present the final strategy. Step 1: Identify the key trends in the data → Step 2: Derive problems based on the trends → Step 3: Present solutions."

🧩 Injecting Patterns with Few-Shot Learning

Few-Shot is a method of teaching AI by showing it "examples." It's showing it "do it this way." It particularly shines in classification or data transformation tasks.

💡 Practical Template (Few-Shot):

[Instruction] Referring to the examples below, proceed with category classification for the new product name. [Example 1] Input: 'Smart Tumbler X10' / Output: 'Device/Home Appliances' [Example 2] Input: 'Eco Vegan Soap' / Output: 'Beauty/Eco-Friendly' [New Input] Input: 'AI-based Financial Report Tool' / Output: ?

Using this structure, AI doesn't just "find an answer"; it "follows the pattern I provided to complete the answer."

3. Structuring Techniques to Force the Desired Output Format (JSON/Markdown)

The most common way to use AI outputs in practice is pasting the results into other systems (e.g., Notion DB, website API). The biggest enemy here is "introduction, conclusion, walls of text."

It's important to force the structure of the output. The most powerful method is requesting a JSON schema.

💡 Practical Template (Forcing JSON):

[Instruction] Write a content plan for the following topic, but the result must follow the following JSON schema. Do not add any other explanations. [Schema] {'title': 'Content Title', 'target_audience': 'Main Target', 'key_message': 'Key Message', 'suggested_format': 'Recommended Format (Card News/Blog)'}

If JSON is too complicated, you can also structure it using Markdown.

## 🚀 [Title]
**🎯 Target:** [Target Definition]
**💡 Key Message:** [The phrase to emphasize most]
---
**[Detailed Components]**
1. [Subtitle 1]: Content summary
2. [Subtitle 2]: Content summary

4. Prompt Engineering: Now Making It a Job Competency

The 5 formulas we've learned so far (role assignment → CoT → Few-Shot → format enforcement) are not so much individual techniques as a methodology for teaching AI the flow of logical thinking.

Recently, advanced architectures like RAG (Retrieval-Augmented Generation) have emerged, and AI performance has increased explosively. But what you must remember is that no matter how outstanding the architecture, the importance of the ability to design "what to ask"—that is, question-design skill—never diminishes.

Prompt engineering is now establishing itself as an essential job competency for people who handle AI well.

✨ Practical Tip: It doesn't have to be perfect from the start. Practice combining the 5 elements above.

1. [Role Assignment] "You are a marketing expert with 10 years of experience."
2. [Goal Presentation] "Create 3 slogans to promote this product."
3. [Constraints] "The slogans must be 15 characters or less and include humor."
4. [Output Format] "Show the results in table (Table) format."

With just this combination, your prompt becomes a work instruction that goes beyond a simple question. Try it right now!