🎨 5 Prompting Patterns That Work Everywhere

1️⃣ Zero-Shot — A Simple Question Without Examples

The most basic form: you provide a task, and the model responds. No examples.

Example: "Translate to Spanish: Welcome to our store"

When to use: simple, straightforward tasks. Translation, summarization, classification of obvious things.

When it does NOT work: custom formats, specific styles, edge cases.

2️⃣ Few-Shot — Providing Format Examples

You show 2-5 examples of the correct response, then ask a new question. The model copies the pattern.

Example:

Classify customer reviews into categories: POSITIVE / NEGATIVE / NEUTRAL.

Example 1: "Fast delivery, great product!" → POSITIVE
Example 2: "The order arrived broken, support is not responding" → NEGATIVE
Example 3: "Bought it, using it, nothing special" → NEUTRAL

Classify: "The price is fine, but the packaging is bad, the product itself is OK"
→

When to use: when you need a specific response format. Classification, data extraction, custom formatting.

Pro tip: 3-5 examples is the sweet spot. Fewer, and the model guesses. More, and it loses focus.

3️⃣ Chain-of-Thought — "Think Out Loud"

You ask the model to reason step-by-step first, then provide the answer. The magic phrase is: "Think step-by-step."

Example:

I own a coffee shop in Kyiv. 50 orders/day, average check is $8.
Rent expense is $1200/month, coffee+milk is $0.80/order, barista salary is $800.
What is my monthly profit, and is it worth expanding to a second location?

Think step-by-step, then provide a conclusion.

When to use: calculations, multi-step analysis, logic problems. Without CoT, the model often "guesses" the final answer and makes mistakes. With CoT, it reasons and gets it right more often.

4️⃣ Role-Play — "You are an experienced X"

You assign an expert role to the model. It begins using the terminology, frameworks, and tone of that profession.

Example:

You are an experienced CFO with 15 years in SaaS companies. I am an early-stage founder,
my MRR is $8k, monthly churn is 6%, monthly growth rate is 12%.
Evaluate the health of my business using three metrics: LTV/CAC, magic number,
and Rule of 40. Give a recommendation: focus on growth or retention?

When to use: when you need an expert perspective. Finance, law, marketing strategy, code review.

Pro tip: combine with CoT: "You are a CFO. Think step-by-step how a CFO would reason about this situation."

5️⃣ Structured Output — Fixed Format

You strictly define the response format. JSON, tables, Markdown with specific sections.

Example:

Analyze this email from a customer: "[email text]"

Return strictly JSON in the format:
{
  "intent": "purchase | complaint | question | other",
  "urgency": "high | medium | low",
  "sentiment": "positive | neutral | negative",
  "suggested_action": "string"
}

No text before or after the JSON.

When to use: when your code / pipeline will parse the response programmatically. JSON, CSV, XML — the model is capable of generating them.

Pro tip: Claude and GPT-4 support "JSON mode"—where it is impossible to get anything other than JSON.

🥋 Battle Combination of All 5

[Role]     You are an experienced sales analyst with a focus on B2B SaaS.
[Few-Shot] Examples of my successful emails:
            1. "{example 1}" → conversion 23%
            2. "{example 2}" → conversion 18%

[CoT]      Analyze the patterns in these emails STEP-BY-STEP:
            - Length
            - Structure  
            - Triggers
            - Call-to-action

[Task]     Generate 3 new cold emails for the lead:
            [lead context]

[Format]   Response structure:
            ## Pattern Analysis
            ## Email 1
            ## Email 2
            ## Email 3
            ## Conversion Forecast

This is already an engineering approach to prompting. Not just "help with an email," but orchestration.

🎬 What's in Lesson 3