Prompt Engineering 101: 5 Core Principles for Getting the Answers You Want from AI
Ever asked an AI chatbot a question and gotten an off-target answer instead of what you wanted? The problem isn’t the AI. It’s how you ask. Prompt engineering is the skill of giving AI precise instructions to get the results you want. In this guide, B2B sales reps and office professionals work step by step through 5 core principles they can apply to their jobs right away.
Before You Start: Set Up Your Practice Environment
Open ChatGPT: In your browser, go to chat.openai.com.
Start a new conversation: Click the "New chat" button at the top left.
Choose a model: From the dropdown at the top, select GPT-4 or GPT-4o. If you’re on the free plan, GPT-3.5 works fine too.
Find the input box: The "Message ChatGPT" text box at the bottom of the screen is where you write your prompts.
💡 The same principles apply to other AI tools such as Claude (claude.ai) and Gemini (gemini.google.com). Whichever AI you use, you can follow this guide as is.
Principle 1: Role Assignment
Why Assign a Role?
When you tell the AI **"You are a ~"**, it answers from that expert’s perspective. Without a role, the AI produces generic, run-of-the-mill answers. Experiments suggest that assigning a role improves the expertise of answers by 40–60% on average.
Step-by-Step Practice
Step 1. Place your cursor in the ChatGPT input box.
Step 2. State the role in the first sentence of your prompt. The format is:
You are a [job title/role] with [experience/area of expertise].
Step 3. Follow the role with your specific request.
Before / After Comparison
❌ Before (no role):
Write an email to send to a customer.
Result: A generic, vague email that reflects nothing about the industry.
✅ After (role assigned):
You are a B2B SaaS sales expert with 10 years of experience. Write an email asking an existing client’s IT manager, who hasn’t been in touch for three months, for a follow-up meeting. Keep the tone polite but not pushy, and the length under 150 words.
Result: An email that naturally uses SaaS industry terminology and is ready to use in real sales work.
Role Assignment Template Collection
Work Situation | Role Assignment Example |
|---|---|
Writing a proposal | You are a solutions sales consultant with 15 years of experience dealing with large-enterprise procurement teams. |
Writing a report | You are a management strategy consultant who came up through McKinsey. |
Customer support | You are a customer success team manager who maintains a 98% CS satisfaction rate. |
Data analysis | You are a marketing data analyst and CRM system administrator. |
💡 The more specific the role, the better. "A senior manager with 10 years of IT solutions B2B sales experience" draws out far more accurate answers than "a salesperson."
Principle 2: Context Setting
Why Does Context Matter?
The AI knows nothing about your situation. Without background information, it fills in the blanks with guesses, and those guesses are likely to be wrong. A prompt with ample context is about 3 times more likely to get you the result you want on the first try than one without.
5 Elements to Include in Your Context
Who — Who is the audience or reader?
What — What deliverable do you need?
Why — What is the purpose of this deliverable?
Where — In what channel or situation will it be used?
Constraints — What rules or limits must be followed?
Step-by-Step Practice
Step 1. Type a role-assignment sentence in the input box. (Apply Principle 1)
Step 2. Below it, write "Background:" or "Situation:" and fill in the five elements above one by one.
Step 3. Press Enter (or the send button) to submit the prompt.
Before / After Comparison
❌ Before (no context):
Write a sales proposal.
Result: A generic template with no clear industry, customer, or product. Takes 30+ minutes to revise.
✅ After (context provided):
You are a sales consultant specializing in cloud security solutions.
Background:
Our company: A cloud security SaaS startup (50 employees, Series B)
Product: "SecureShield," a real-time threat detection platform (subscriptions from KRW 5 million (≈ US$3,740) per month)
Client: The IT security team at a mid-sized manufacturer with 3,000 employees
Client challenge: Recently hit by a ransomware attack and feeling the limits of their existing on-premises security solution
Competitor: Currently using Company P's on-premises solution
Purpose: A follow-up proposal sent after the first meeting (to be reviewed by the decision-maker, the CISO)
Write a draft two-page sales proposal for the situation above. Start with the client's pain points, then include what differentiates our solution, an ROI example, and the next steps.
Result: A proposal tailored precisely to the client's situation. Revision time: under 5 minutes.
💡 If your background information runs long, organize it with bullet points (-). AI understands structured information more accurately.
Principle 3: Output Format Specification
Why should you specify a format?
Even with identical content, how useful the result is depends entirely on its format: table, list, email, or report. Without a specified format, AI tends to answer in long prose, which is hard to copy and use right away. Specifying a format cuts follow-up editing time by an average of 70%.
Examples of output formats you can specify
Format | Prompt keyword |
|---|---|
Table | "Organize this in a table" |
Numbered list | "Write this as a numbered list" |
"Write this as an email I can send right away" | |
JSON/CSV | "Output this in JSON format" |
Slide outline | "Separate the title and content for each PowerPoint slide" |
Dialogue script | "Write this as a conversation between a salesperson and a customer" |
Step-by-Step Practice
Step 1. State the output format you want at the end of your prompt.
Step 2. If needed, also specify the length, language, and tone.
Step 3. Showing a sample of the output you want raises accuracy even further.
Real-World Prompt Example
You are a B2B marketing expert.
Compare and analyze the CRM solutions of the following 3 competitors:
Salesforce
HubSpot
Pipedrive
Output format:
Write it as a comparison table
Comparison criteria: price range, key features, pros, cons, recommended company size
Below the table, write a recommendation of “the best fit for our team (a startup with 5 salespeople)” in 100 words or fewer
English, polite and professional tone
Result: A clean comparison table is generated instantly, ready to use as material for a team meeting.
Before / After Comparison
❌ Before: “Compare CRM solutions” → long paragraphs of text, 5 minutes to grasp the key points
✅ After: A prompt with a specified format → an organized comparison table, 10 seconds to grasp the key points
💡 If you say “Write it as a markdown table,” ChatGPT will build a clean table. Copy it and paste it into Notion or Google Docs, and the table format is preserved.
Principle 4: Guide Step-by-Step Reasoning
Why should you make the AI think step by step?
If you throw a complex question at the AI all at once, it may skip the intermediate steps and reach an inaccurate conclusion. Research (Kojima et al., 2022) shows that adding a single sentence, “Think step by step,” can improve logical accuracy by up to 40%.
Step-by-Step Practice
Step 1. Apply this principle when requesting complex analysis, calculations, or strategy development.
Step 2. Add one of the following phrases to your prompt:
“Think step by step, then reach a conclusion.”
“First show your analysis, then give the final answer.”
“Proceed in this order: Step 1: ~, Step 2: ~, Step 3: ~.”
Step 3. If possible, design the reasoning steps yourself and include them in the prompt.
Real-World Prompt Example
You are a B2B sales strategy consultant.
Our company (an HR automation SaaS) plans to enter the enterprise market next year. We currently have 200 small and mid-sized business customers, with an average contract value of KRW 1 million (≈ US$750) per month. Our goal is to double revenue next year.
Please build a sales strategy by following these steps:
Step 1: Analyze the current situation (strengths, weaknesses, opportunities, threats — SWOT)
Step 2: Analyze the enterprise customer’s purchase decision process
Step 3: Define the ideal customer profile (ICP)
Step 4: Create a concrete sales action plan (by quarter)
Step 5: Define key KPIs and how to measure performance
Please write up each step in detail, in order.
Result: A well-structured strategy document is generated, with each step logically connected to the next.
Before / After Comparison
❌ Before: “Draft a sales strategy” → 5 lines of superficial generalities
✅ After: Step-by-step reasoning → an actionable strategy document, from the SWOT analysis to a quarterly action plan (2–3 pages at A4/letter size)
💡 For any request involving numbers (ROI calculations, revenue forecasts, etc.), always add “Show your calculations step by step.” That makes it easy to spot and fix the AI’s calculation errors.
Principle 5: Iterative Refinement
Why shouldn’t you expect a perfect answer on the first try?
The heart of prompt engineering is iterative refinement through conversation. Even experts rarely get a perfect result from a single prompt. Typically, 2–4 rounds of revision get you to a result you’re happy with. The key is to give the AI specific feedback.
Types of Effective Feedback
Feedback type | Example phrase |
|---|---|
Change the tone | “Rewrite this in a more formal tone.” |
Adjust the length | “Cut the current text in half. Keep only the key points.” |
Edit a specific part | “In the second paragraph, replace the ROI section with a concrete numerical example.” |
Change the perspective | “Rewrite this from a CFO’s perspective instead of a CTO’s.” |
Add something | “Add three ways we differ from our competitors.” |
Change the format | “Convert this from an email into a slide outline.” |
Step-by-Step Practice
Step 1. Send a prompt that applies Principles 1–4 and review the result.
Step 2. Separate the parts you like from the parts that need changes.
Step 3. Enter a follow-up prompt in the same conversation. Don’t start a new chat — the AI remembers the earlier context.
Step 4. Make your revision requests specific.
Iterative improvement in practice
First request (applies all of Principles 1–4):
You are a B2B SaaS sales manager with 10 years of experience. Write an email to encourage a paid conversion three days before the free trial ends for our project management solution, “TaskFlow.” The recipient is a team lead at a 50-person marketing agency. The tone should feel like a helpful partner. Keep it under 150 words and write it in polite, professional English.
Second round of feedback:
Good. Please revise the following:
The first sentence sounds too salesy. Change it to open with what the customer achieved during the trial.
In the pricing section, add the benefit “20% off with annual billing.”
Change the CTA (Call to Action) to “a quick 15-minute call.”
Third round of feedback:
Almost perfect. Lastly, suggest three more subject line options, in a style that gets high open rates.
Result: in three rounds, a sales email ready to send. Total time: about 5 minutes.
Before / After Comparison
❌ Before: Unhappy with the result, you start a new chat → re-explain the context from scratch every time, wasting time
✅ After: Improve iteratively with specific feedback in the same conversation → final result within 3 rounds, saving 80% of the time
💡 Make use of the conversation history on the left side of ChatGPT. If you rename conversations that contain well-crafted prompts (click the conversation title → pencil icon → rename), you can reuse them later for similar tasks.
A master prompt template that combines all five principles
Copy the template below and edit only the text inside the [square brackets] to fit your situation.
[Role] You are a [job title] specializing in [field], with [N years] of experience.
[Context]
Our company: [company description]
Situation: [current situation/background]
Audience: [who will read the output]
Goal: [what you want this output to achieve]
[Request] Please write a [deliverable type] that fits the situation above.
[Output format]
Format: [table/list/email/report, etc.]
Length: [word count or page count]
Tone: [formal/casual/professional, etc.]
Language: [Korean/English]
[Step-by-step thinking] Please write in this order:
[Step 1]
[Step 2]
[Step 3]
3 Practice Exercises
Open ChatGPT right now and try the exercises below:
Exercise 1: Write a cold email (Difficulty: ★☆☆)
Apply Principle 1 (Role) + Principle 2 (Context) + Principle 3 (Format)
to write a first email to a prospect you have never met.
Exercise 2: Analyze quarterly results (Difficulty: ★★☆)
Focus on Principle 4 (Step-by-step thinking)
Give the AI some hypothetical quarterly sales data and ask it to write up “cause analysis → insights → improvement plan,” in that order.
Exercise 3: Complete a proposal (Difficulty: ★★★)
Apply all 5 principles
to write a proposal you actually need at work, and go through at least 2 rounds of iterative refinement (Principle 5).
Top 5 Common Mistakes and How to Fix Them
Rank | Mistake | Fix |
|---|---|---|
1 | The prompt is too short and vague | Use the 5 context elements from Principle 2 as a checklist |
2 | Asking for several things at once | Split complex requests into 2–3 prompts |
3 | Starting a new chat when the result disappoints | Refine iteratively in the same chat with specific feedback (Principle 5) |
4 | Using AI answers as-is without checking | Always fact-check numbers, proper nouns, and dates |
5 | Writing every prompt from scratch | Save your master template in Notes or Notion and reuse it |
Wrap-Up: Your Action Plan Starting Today
Today: Save the master prompt template to your favorites or notes.
This week: Try at least 1 of the 3 practice exercises.
This month: Apply the 5 principles to real work emails, reports, and proposals.
Goal: In 4 weeks, cut your AI-assisted document writing time by 50% or more.
💡 Prompt engineering is about practice, not theory. Rather than agonizing over the perfect prompt, the fastest way to learn is to write one and keep refining it. Open ChatGPT right now and start with the first practice exercise!