How to Structure ChatGPT Prompts: A 4-Part Template plus CoT and Few-Shot

ChatGPT 100% Utilization Guide: A Universal Prompt Structuring Formula Even Beginners Can Use Immediately

"Can't I just ask the AI a question?"

Have you ever thought that and thrown a vague question at ChatGPT? Questions like "Give me some marketing ideas" or "Write a draft of the report" are a lot like asking a chef to just "make something tasty."

AI is an all-purpose engine with enormous potential, but if you don't provide a specific blueprint for how to run that engine, the output often falls short of expectations.

This article is for non-specialists, planners, and marketers who want to apply AI to their work but feel stuck on what to ask. We'll completely break down the concept of "prompt engineering" that felt complicated, and share a universal structured formula anyone can copy and paste starting today.

Why isn't just 'asking questions' enough? Understanding the importance of prompts

Prompt engineering is the technique of designing and refining questions (inputs) so that an AI model (LLM) produces the optimal desired output.

In simple terms, it's the process of systematizing the skill of asking AI questions. The core isn't just listing keywords—it's assigning the AI a role (Persona), providing background knowledge (Context), and clearly requesting an output format (Format).

💡 Bad Prompt vs. Good Prompt: A Comparative Analysis

Theory alone may not click, so let's compare 'bad prompts' and 'good prompts' through three real work scenarios.

As you can see, a good prompt is like a clear instruction manual that tells the AI not only what to do, but also from whose perspective, in what order, and in what form.

🚀 The 4-Step Prompt Structure Formula That Elicits the Best Answers (Universal Template)

Now I'll introduce a 4-step structure that applies to every situation and can be used immediately in practice. Remember this order and your confidence in using AI will rise 100%.

📝 Universal Prompt Template You Can Copy and Use Right Away

[ (Persona)]
 [ ]. [ ].

[ (Context)]
 [ , , ]. .

[ when (Instruction)]
 :
1. [ ]
2. [ ]
3. [ /]

[ set (Format)]
Result when [ : , , 500 ] .

Key points for each step:

1. Role Assignment (Persona): Start with "You are ~." to define the AI's identity. (e.g., professional copywriter, business consultant, friendly tutor)
2. Context Provision (Context): Provide all information the AI should reference at once. (e.g., product specs, market trend reports, customer feedback)
3. Specific Instructions (Instruction): Clearly list what to do and how to do it in verb form.
4. Output Format Specification (Format): Specify the form of the output. (e.g., must be a table, use titles and subtitles, present only 3 options)

🧠 Advanced Techniques for Eliciting AI's Thought Process (CoT & Few-Shot)

Once you've mastered the basic structure, it's time to look at advanced techniques for controlling the depth of the AI's thinking.

1. Applying CoT (Chain-of-Thought)

If you simply demand an answer, AI tends to produce only the 'conclusion.' For more complex problems, you need to ask it to show the thinking process. That's CoT.

[CoT Application Example: Establishing a Market Entry Strategy]

(Add to instructions) "Before presenting the answer, you must go through the following 3-step thought process. Step 1: Set core assumptions → Step 2: Create a strengths/weaknesses matrix vs. competitors → Step 3: Based on this analysis, derive the single most realistic entry strategy."

By requesting a step-by-step thought process like this, AI provides in-depth analysis with logical grounds instead of superficial answers.

2. Few-Shot Learning (Providing Examples)

If the style or format of the desired output is very particular, providing a few examples is the most powerful approach.

• Example: "The following is a summary example for [topic]. [Example 1], [Example 2]. Now please summarize [new topic] in the same format."

By providing examples like this, AI learns the pattern and generates output matching the desired format.

📌 Summary:

If you construct prompts with these 4 steps in mind, you can use AI not as a simple search engine, but as the best professional assistant working alongside you.

When the Template Still Gives Unexpected Answers

If results still vary with a structured prompt, the cause is usually one of the following.

Causes and how to check

• Missing context: audience, purpose, or length is missing. Check that every slot in the template is actually filled.
• Conflicting instructions: requests like "summarize briefly" and "cover every item in detail" appear together. State the priority (e.g., "the length limit comes first").
• Factual questions: questions needing current information or internal material aren't solved by prompt structure. Paste the material in or ask for sources, and check results against the original.

A reproducible comparison

1. Run the same question three times each with the old and revised prompts.
2. Mark results in a table against criteria set in advance (format followed, required items included, factual errors).
3. Adopt the revision into the template only if it meets the criteria all three times.

Cautions

• Don't enter personal data or company secrets, or check the service's data-training settings first.
• Always verify numbers, dates, and quotes against the original. Structured prompts stabilize format but don't guarantee factual accuracy.