The Ultimate Prompt Engineering Cheat Sheet for AI Users

📌 How to Use This Guide

This is a reference document, not a tutorial. If you want the deep explanations behind every technique here, the other posts in this series cover them in full. This guide is for the moment when you are about to send a prompt and want to quickly check whether you have covered everything that matters.

Bookmark it. Come back to it. Use it the way you would use a style guide or a code review checklist — not as a substitute for understanding, but as a reminder when you are moving fast and want to catch the things you might otherwise miss.

🧱 The 7 Core Elements of a Strong Prompt

Not every prompt needs all seven. But every prompt you write should be checked against this list. Missing elements are almost always the reason an output disappoints.

1. ROLE
   Who should the AI be?
   Example: "You are a senior product manager with experience in 
             B2B SaaS companies."

2. TASK
   What specific action should it take?
   Use precise verbs: rewrite, extract, compare, summarize, 
   classify, generate, evaluate — not just "help" or "do something."

3. CONTEXT
   What does it need to know about your situation?
   Audience, purpose, prior work, constraints, domain.

4. FORMAT
   How should the output be structured?
   Bullet points, table, numbered list, JSON, paragraphs, 
   word count, number of items, heading structure.

5. TONE
   How should it sound?
   Formal, conversational, technical, empathetic, direct, 
   persuasive, neutral. Name the register explicitly.

6. CONSTRAINTS
   What must it avoid?
   Jargon, caveats, specific phrases, certain formats, 
   competitor mentions, topics out of scope.

7. EXAMPLES (optional but powerful)
   Show what good output looks like.
   One or two examples shift output quality more than 
   almost any other single addition.

⚡ Technique Cheat Codes

For Better Reasoning

ADD: "Think step by step before giving your final answer."
ADD: "Explain your reasoning at each stage."
ADD: "Consider at least two alternative approaches before choosing one."
ADD: "What assumptions are you making? State them before continuing."

For More Accurate Outputs

ADD: "If you are uncertain about any part of this, say so explicitly."
ADD: "Do not fill gaps with assumptions — ask me instead."
ADD: "Base your answer only on the information I have provided."
ADD: "If this question requires knowledge beyond your training, say so."

For Consistent Format

ADD: "Return your answer as valid JSON only. No explanation outside 
      the JSON object."
ADD: "Use exactly this structure: [paste your template]"
ADD: "Every response in this conversation must follow this format: 
      [your format spec]"
ADD: "Stop generating after the closing bracket / after the last 
      bullet point / after exactly N sentences."

For Better Creative Output

ADD: "Do not use any of these phrases: [list clichés to avoid]"
ADD: "Write as if this will be published under your byline — 
      not as AI-generated content."
ADD: "Generate three versions with different tones and I will 
      choose the best one."
ADD: "The first draft should be deliberately unconventional. 
      I will ask for a safer version if needed."

For Handling Long Tasks

ADD: "Complete only Step 1 and wait for my confirmation 
      before proceeding."
ADD: "After each section, ask me if I want to continue or adjust."
ADD: "If this task will require output longer than 500 words, 
      ask me to confirm scope before proceeding."

🔧 Ready-to-Use Prompt Templates

Universal Writing Template

Role: [Who the AI should be]
Audience: [Who will read this]
Task: [Specific writing action — rewrite, draft, summarize, edit]
Format: [Structure, length, heading style]
Tone: [Register and voice]
Constraints: [What to avoid]
Context: [Relevant background]

[Paste your content or describe the topic]

Email Drafting Template

Draft an email with the following:

Purpose: [One sentence on what this email must accomplish]
Recipient: [Who they are, your relationship, what they care about]
Tone: [Professional/warm/urgent/casual]
Length: [Under X words / 3 paragraphs max / one screen]
Key points to include: [List]
Things to avoid: [e.g., don't sound pushy, don't mention price]
Subject line: [Write 3 options with different angles]

Draft the email body first, then the subject lines.

Summarization Template

Summarize the following [article/document/conversation] for 
[audience description].

Format:
- 2-sentence overview
- [N] key points as bullets
- 1 sentence on the most important implication

Length: Under [X] words total.
Avoid: Jargon, repetition between sections, personal opinion.

[Paste content here]

Code Review Template

Review the following [language] code. Provide feedback on:
1. Correctness — any bugs or unhandled edge cases?
2. Readability — naming, structure, comments
3. Performance — anything obviously inefficient?
4. Security — any obvious vulnerabilities?

For each issue: quote the specific line, explain the problem, 
show the fix.

Do not rewrite the entire function unless critical. 
Prioritize issues by severity: Critical / Warning / Suggestion.

[Paste code]

Research Synthesis Template

Synthesize the following sources into a structured briefing.

Sections:
1. Key findings (3-5 bullets)
2. Points of agreement across sources
3. Points of contradiction or uncertainty
4. Most significant implication for [your context]
5. What is still unknown or requires further research

Be explicit about when you are drawing inferences versus 
stating what the sources actually say. Label accordingly.

[Paste sources]

Decision Support Template

Help me think through this decision: [describe decision]

Context: [key facts, constraints, what you already know]

Provide:
1. A structured pros/cons analysis
2. The 3 most important questions I should answer before deciding
3. The most likely failure mode of each option
4. One non-obvious consideration I might be missing

Do NOT give me a final recommendation — I want to make 
the decision myself with better information.

🚫 The Most Common Prompt Mistakes and Their Fixes

MISTAKE: "Help me with my email."
FIX: "Rewrite the following email to be more concise and 
      confident. Target: under 100 words. Keep all key information."

MISTAKE: "Write a blog post about marketing."
FIX: "Write the introduction paragraph (100-120 words) for a blog 
      post titled: [your specific title]. Audience: [specific reader]. 
      Tone: [register]. Hook with: [specific angle, not a generic opener]."

MISTAKE: "Summarize this."
FIX: "Summarize the following for a C-suite executive who has 
      90 seconds to read it. Format: 3-sentence overview + 
      4 bullet key takeaways. Under 150 words total."

MISTAKE: "Make this better."
FIX: "The current version is too formal and uses jargon our 
      audience will not understand. Rewrite it for a small 
      business owner with no technical background. Keep the 
      same information but replace all technical terms with 
      plain language. Preserve the three key recommendations."

MISTAKE: Starting with a long monologue, then asking the question 
         in the final sentence.
FIX: State the task first. Context second. Details third.

MISTAKE: Accepting the first output without iteration.
FIX: Treat every first output as a draft. Identify the specific 
     gap (tone? depth? format? accuracy?) and address that 
     specific thing in your follow-up.

🎯 Power Phrases That Instantly Improve Output

For Precision

• ✅ "Be specific — no generalities."
• ✅ "Use concrete examples, not abstract principles."
• ✅ "Every claim must be supported by something in the provided material."
• ✅ "Name things, do not describe them vaguely."

For Conciseness

• ✅ "Cut every word that does not earn its place."
• ✅ "If a sentence repeats what the previous one said, delete it."
• ✅ "Maximum [N] words. If it cannot fit, cut — do not extend."
• ✅ "No filler openings. Start with the main point."

For Avoiding Generic Output

• ✅ "Do not use: 'In today's world', 'It is important to note', 'As an AI', or 'Certainly'."
• ✅ "Write from a specific perspective, not a neutral overview."
• ✅ "Take a position. Do not hedge every sentence."
• ✅ "The reader already knows the basics. Skip the introduction and get to the insight."

For Uncertainty and Accuracy

• ✅ "Distinguish clearly between what you know and what you infer."
• ✅ "If you are less than confident about a fact, say so."
• ✅ "Do not fill gaps — note them."
• ✅ "If this requires real-time data you do not have, say so rather than estimating."

🔄 The Iteration Checklist

When an output is not right, run through this before re-prompting:

□ Is the task specific enough? (vague verbs = vague output)
□ Have I defined the audience?
□ Have I specified the format?
□ Have I set the tone?
□ Have I told it what to avoid?
□ Have I given context it would not already know?
□ Would adding one example clarify what I want?
□ Am I trying to do too many things in one prompt? 
  (If yes — split into steps)
□ Is the gap in the output about tone, depth, accuracy, 
  format, or focus? (Address that specific gap in the follow-up)

📐 Prompting by Task Type — Quick Reference

WRITING TASKS
└─ Always specify: audience, tone, format, length, what to avoid
└─ For drafts: use role + voice description + audience
└─ For editing: be specific about WHAT to change and WHY

RESEARCH TASKS
└─ Always specify: what to synthesize, level of detail, 
                    format for findings
└─ Add: "Distinguish facts from inferences"
└─ Add: "Note when information is absent rather than speculating"

CODING TASKS
└─ Always specify: language, libraries allowed, constraints, 
                    whether to include comments, example usage
└─ For debugging: paste the error AND the relevant code
└─ For review: specify which dimensions to evaluate

ANALYSIS TASKS
└─ Define: what question the analysis answers, 
             what criteria to use, what to ignore
└─ Specify: depth (surface scan vs deep dive)
└─ Add: "Prioritize findings by importance, not by 
          order of appearance in the source"

CREATIVE TASKS
└─ Describe the voice, not just the topic
└─ Give a specific angle rather than an open topic
└─ List phrases, structures, and styles to avoid
└─ Ask for multiple versions when the brief is ambiguous

🗝️ The Three Questions to Ask Before Every Prompt

These three questions, asked quickly before you send any prompt, catch the majority of issues that lead to disappointing output:

• ❓ What does a smart assistant who knows nothing about my situation need to know to answer this well? — If the answer is 'more than what I wrote,' add it.
• ❓ What format would actually be most useful for me right now? — If the format is obvious to you, state it. Do not let the model guess.
• ❓ What would make this output unusable — and have I told it to avoid that? — Negative constraints are as important as positive ones.

Prompt engineering improves with practice, but these three questions accelerate that improvement faster than almost anything else. They force the habit of reading your prompt as the model would read it — from a position of knowing nothing about your intent beyond what you wrote.

That shift in perspective is the foundation of everything in this guide. Apply it consistently, and the quality of what you get back from any AI tool will reflect it.