Hey
I will be real with you about something I have never fully admitted.
The first time I heard the phrase "prompt engineering," I thought it was for software engineers.
Like, the word engineering is right there. I pictured someone in a dark room with 4 monitors writing code just to talk to an AI. I thought, this is not for me. I run a newsletter. I am not an engineer. I am not even close.
So I did what most people do. I ignored it.
And then the AI Twitter guys made it worse.
They would post these massive, complicated prompts with XML tags and role hierarchies and "chain of thought reasoning" and use words like "token optimization" and "latent space." I would read those threads and feel dumber with every scroll. They made it sound like rocket science. Like you needed a PhD just to write a sentence to an AI.
I stayed scared of it for months.
And here is what I know now, sitting on the other side of it: it was never that complicated. The scary people just wanted to sound smart. That is it.
Prompt engineering is just knowing how to talk to Claude properly.
That is the whole thing.
And right now, most people still do not know how to do it. They open Claude, type "write something about marketing," get a generic paragraph, and blame the tool. They think Claude is mediocre. But Claude is not mediocre. Their prompt is.
The model does not change. The subscription does not change. The only variable is how clearly you communicate with it.
So today I am breaking the whole thing down. Zero to expert. No jargon. No gatekeeping.
Level 1: Stop Typing Like a Robot
Most people write prompts like this:
_"Write me a blog post about AI trends."_
That is like walking into a restaurant and saying "give me food." You will get something. But it will not be what you wanted.
Here is what a real prompt looks like:
_"Write a 1,500-word blog post about the three biggest AI trends for 2026. My audience is small business owners who are not technical. Use simple language, short paragraphs, and open with a surprising stat. Do not use words like 'leverage,' 'synergy,' or 'paradigm.' End with three things they can do this week."_
Same Claude. Same subscription. Completely different output.
The rule is simple: every detail you leave out is something Claude has to guess. And Claude's guesses are not your preferences.
Level 2: The Five Things Every Good Prompt Needs
Think of this as your checklist. Every time you sit down to prompt Claude, run through these five things:
1. Role — Tell Claude who it is. "You are a senior email marketer with 10 years of experience in ecommerce" gives Claude a voice. Without a role, Claude defaults to generic helpful assistant, which is nobody's ideal.
2. Context — What does Claude need to know? Your audience. Your goal. Your industry. The more relevant context you give, the more specific the output.
3. Task — Not approximately what you want. Exactly. "Analyze this" is vague. "Identify the top 3 risks in this contract and suggest specific edits to fix each one" is exact.
4. Format — Do you want a list? A paragraph? An email? A table? If you do not say, Claude picks. Sometimes it picks wrong.
5. Constraints — What should Claude NOT do? This is the most underrated part. "Do not use passive voice. Do not add unnecessary caveats. Do not start sentences with 'It is important to note.'" Six to ten negative constraints will make your output sound less like AI and more like an actual human wrote it.
Use this checklist every single time.
Level 3: The Techniques That Separate Beginners From Experts
Give examples. One example does more than ten paragraphs of instructions. Do not describe what you want in abstract terms. Show Claude a concrete example and say "write more like this."
Break big tasks into steps. Do not ask Claude to research, analyze, outline, and write a full report in one go. Do it in four prompts. Each step builds on the last and the quality compounds.
Use XML tags. This sounds scary but it is just organized writing. Instead of one long paragraph of instructions, you separate them like this:
```
You are a newsletter writer for a tech-curious audience
Write a 600-word issue about Claude's new features
No jargon. No bullet overload. Keep it conversational.
```
Claude was trained on structured prompts. This format removes ambiguity and produces more consistent output.
Ask Claude to evaluate itself. At the end of any prompt, add: "After writing, rate your response 1-10 on clarity and usefulness. If any score is below 8, improve it and show only the final version." Claude catches its own weak spots before you even see them.
Give specific feedback, not vague feedback. Do not say "make it better." Say "the opening is too generic, replace it with a personal story. Section 2 is too long, cut it by half. The ending should be a declaration, not a question." Specific feedback gets specific improvements.
##### **The Master Template (Save This)**
Here is the one prompt structure worth saving and reusing for everything:
```
[Who Claude is for this task]
[Your audience, goal, situation]
[Exactly what you want]
[One or two examples of the output style you want]
[Structure of the output]
[What to avoid]
[Ask Claude to self-check before delivering]
```
Fill this once per task type. Save it. Reuse it. The people doing this are moving at a completely different speed from everyone else.
Level 4: 5 Ready-To-Use Prompts You Can Copy Right Now
Most people read about prompting and still do not know what to actually type.
So here are five prompts you can copy, fill in the brackets, and use today.
The Analysis Prompt
_"You are a [domain] expert with 15 years of experience. Analyze [subject] and identify the 3 most significant insights. For each one, give me: (1) a clear statement of what it is, (2) specific evidence supporting it, (3) why it matters for [your audience], and (4) one recommended action. Use specific numbers wherever possible. Do not hedge or add unnecessary caveats."_
Use this when you need Claude to think deeply about something, not just summarise it.
The Writing Prompt
_"You are a professional writer. Write a [format] about [topic] for [audience]. Open with a hook that challenges a common assumption. Use short paragraphs. Every sentence should either teach something, prove something, or move the reader forward. No filler phrases, no corporate jargon, no passive voice. Target [word count] words."_
Use this for any content you are creating. Newsletters, posts, emails, scripts.
The Decision Prompt
_"I need to decide between [Option A] and [Option B]. Here is my situation: [context]. Analyze each option across these criteria: [criteria 1], [criteria 2], [criteria 3]. For each criterion, rate each option 1-10 and explain the score in one sentence. Then give me your overall recommendation with a confidence level (high / medium / low) and tell me the one piece of information that would most change your recommendation."_
Use this when you are stuck on a decision and going in circles.
The Problem-Solving Prompt
_"I am experiencing [problem]. Here is what I have already tried: [attempts]. Here is what I know about the root cause: [knowledge]. Diagnose the most likely cause. Propose three solutions ranked by likelihood of success. For each, estimate the effort required and the probability it will work. Recommend the best path forward."_
Use this when you are frustrated with something and need structured thinking, not sympathy.
The Feedback Prompt
_"Review [my work] against these criteria: [criteria]. For each criterion, rate it 1-10 and explain specifically what works and what does not. Identify the single highest-impact improvement I could make. Rewrite the weakest section to show me what excellent looks like. Be direct. I prefer harsh truth over gentle encouragement."_
Use this when you want real feedback, not validation.
Save these five. Customize the brackets every time. These will cover 90% of everything you do with Claude.
My Final Words
Here is what I wish someone had told me when I was scared of that phrase two words long.
Prompt engineering is not about memorizing tricks. It is about clarity of thought.
The reason most people get bad output from Claude is not because the model is weak. It is because they have not figured out what they actually want. The prompt is just the reflection of your thinking. When your thinking is clear, the prompt is clear, and the output is good.
The AI Twitter guys who made it sound like rocket science were not teaching you. They were performing. Do not let them make you feel like this is not for you.
It is for you. It always was.
Start with the five-part checklist on your next prompt. One prompt. That is all. You will see the difference immediately and you will never go back to typing two-word instructions and hoping for the best.
Stay curious, talk to you tomorrow.