The structured thinking behind great prompts — and what to do when you're blank and don't know where to start.
It's not about using magic words. It's not about prompt formulas you found on YouTube.
It comes down to one thing: does Claude have enough context to do the job well?
Context engineering is the skill of figuring out what Claude needs to know — before you type anything.
Before you write a single word, ask yourself:
"If I was a brilliant new employee assigned this task for the first time, what would I need to know to nail it?"
Answer that question, then put those answers in your prompt.
That's it. That's context engineering.
Not every prompt needs all 5 — but these are the things that matter:
Tell Claude what role it's playing.
"You are an expert social media writer for a healthcare staffing company."
This isn't just flavor. It shapes how Claude thinks, what vocabulary it uses, and what level of depth it goes to.
Be specific about the output you want.
❌ "Write a post."
✅ "Write a LinkedIn post, 150-200 words, promoting a travel nurse opportunity in Dallas."
Background information that changes the output.
What must NOT be in the output.
"Don't use medical jargon. Don't sound salesy. No hashtags. Keep it under 200 words."
How do you want the output structured?
"Give me a subject line, then the email body in 3 short paragraphs, then a CTA."
Without context engineering:
"Write an email for InSync."
With context engineering:
"You are a warm, mentor-like career coach writing for InSync Healthcare Staffing. Write a 200-word email to an RN who just applied for a travel nurse position. The goal is to make them feel seen and excited, and to book a discovery call at calendly.com/insync-discovery. Tone: encouraging, personal, human. No jargon, no corporate language. End with one clear call to action."
Same task. Completely different output quality.
This happens to everyone. You know something needs to get done but you can't form a prompt.
The "ask Claude what to ask" trick:
"I need to [accomplish X]. I'm not sure exactly how to ask for this. What questions should I answer so you can help me best?"
Claude will ask you 3-5 clarifying questions. Answer them. Then say "now do it."
Another version:
"I need to create [deliverable] for [client/project]. Before you start, tell me what information you'd need to do this really well."
This flips the script — instead of you trying to figure out what context to give, Claude tells you what it needs.
If you've got a piece of work you like — a great email, a well-written post, a clear report — you can reverse engineer a prompt from it.
"Here is an example of exactly the kind of [email / post / report] I want. Study the style, tone, structure, and length. Then write me a new one about [new topic], following the same style."
Claude will analyze the example and replicate the pattern. This is one of the fastest ways to get consistent output — give it an example instead of trying to describe what you want.
When you're stuck, start here:
For any writing task:
"I need to write [thing]. My audience is [who]. My goal is [what I want them to do or feel]. My tone should be [describe]. Here is any relevant context: [paste it]. Write a first draft."
For client research:
"I'm going to tell you about a client. After I do, give me: 3 content angles for social media, 3 subject lines for email campaigns, and 2 ideas for lead magnets. Ready? Here's the client: [describe]."
For a task you've never done before:
"I need to [task]. I've never done this before. Walk me through what the ideal approach looks like, then ask me for any information you need before we start."
For editing something:
"Here is a draft I wrote. Improve it for clarity and impact. Keep my voice. Tell me what you changed and why. [paste draft]."
For when you got a bad first result:
"That's not quite right. What I actually want is [describe the difference]. Try again."
For ongoing work, your context builds over time:
The more of this you have pre-loaded, the less you have to explain each time.
A sidekick who has done the work in Module 2 (knowledge base + skills) is already halfway to a great prompt before they type a single word.
Take one task you do repeatedly — a weekly report, a client email, a social post — and write the full context-engineered prompt for it using the 5 elements above.
Then save it as a skill. You'll never type it from scratch again.
Next module: Build your AI knowledge base — the permanent foundation that makes every prompt better.
HivePowered AI — AI Like a Pro Training