Prompt Engineering Cheat Sheet

📐 THE CRAFT FRAMEWORK

C

Context

Background & situation

R

Role

Who AI should be

A

Action

What to do

F

Format

Output structure

T

Tone

Voice & style

❌ WEAK

"Help me with marketing"

✅ STRONG (using CRAFT)

[C] I run a B2B SaaS startup targeting SMBs
[R] Act as a growth marketing strategist
[A] Create a 30-day content calendar
[F] Table with: date, platform, topic, CTA
[T] Professional but approachable

🎯 CORE TECHNIQUES

Zero-Shot

Direct instruction, no examples needed

"Translate to Spanish: Hello world"

Few-Shot

Provide 2-5 examples to show pattern

"Happy → Positive, Sad → Negative, OK → ?"

Chain-of-Thought

Ask for step-by-step reasoning

"Solve this. Think step by step."

Role/Persona

Assign expertise to shape response

"You are a senior data scientist..."

Structured Output

Specify exact format (JSON, table, etc.)

"Respond as JSON with keys: summary, action_items"

🚀 COPY-PASTE QUICK WINS

Better Explanations

"Explain [X] as if I'm a smart 12-year-old. Use an analogy."

Improve Any Text

"Rewrite this to be clearer, more concise, and more professional: [text]"

Get Alternatives

"Give me 5 different approaches to [problem]. Rank by effort vs impact."

Challenge My Thinking

"What are the 3 strongest counterarguments to this position?"

Extract Insights

"Analyze this and give me: 3 key insights, 2 risks, 1 recommendation."

🔬 EVALUATE YOUR PROMPTING

🎯

Clarity

Was I specific?

📚

Context

Enough background?

📐

Structure

Format specified?

🔄

Iteration

Did I refine?

🧠

Critical

Did I verify?

⚡

Efficient

Minimal turns?

🧩 POWER PATTERNS

📋 "First, X. Then, Y. Finally, Z." — Sequential tasks

🎭 "Act as [expert] with [years] experience in [field]"

📊 "Present as a table with columns: X, Y, Z"

🚫 "Do NOT include [X]. Focus only on [Y]"

📏 "Keep it under [N] words/bullet points"

🎚️ "On a scale of 1-10, rate [X] and explain why"

⚖️ "Give pros and cons. Recommend one with reasoning"

🔌 API ESSENTIALS

model

Which AI: "claude-sonnet-4-5-20250514"

messages

[{"role": "user", "content": "..."}]

max_tokens

Response length limit (1024 ≈ 750 words)

system

Persistent instructions/persona

temperature

0 = focused, 1 = creative

stop_sequences

Where to stop generating

import anthropic
client = anthropic.Anthropic()
message = client.messages.create(model="claude-sonnet-4-5-20250514", max_tokens=1024, messages=[{"role": "user", "content": "Your prompt"}])
                

⚠️ AI LIMITATIONS — ALWAYS REMEMBER

Hallucinations: AI can confidently state false information. Always verify facts.

Math errors: LLMs struggle with arithmetic. Double-check calculations.

Outdated info: Training data has a cutoff. Ask about recent events with caution.

Bias: AI reflects biases in training data. Consider diverse perspectives.

⚡ Prompt Engineering Cheat Sheet