Here's the 3-part structure that actually works:
Part 1: Context (.md) files.
Part 2: A real instruction block.
Part 3: "Don't start yet, ask me questions first."
That's it. Most people skip all three.
Here's what each one fixes:
1. Context files
↳ Open a doc, name it "VOICE & TONE"
↳ Add 3 to 5 examples of work you'd want to replicate
↳ Open a second doc: "AUDIENCE AVATAR"
↳ Add who you're writing for, their pain points, what they care about
↳ Save both as .md files and you only have to do this task once
↳ Upload before you type a single word
↳ Now Claude knows who you are and who you're talking to
2. A real instruction block
↳ Assign a role: "Act as a LinkedIn ghostwriter for solopreneurs"
↳ Tell it to read the attached files completely first
↳ State the task in one sentence
↳ Set the format and length (3 post options, under 150 words each)
↳ Tell it to refine once after you pick a winner
↳ That's a prompt. Not 500 words. Not one vague sentence.
3. "Ask me questions first"
↳ Add this line before the task runs:
↳ "Don't start yet. Ask me clarifying questions first."
↳ Claude comes back with 3 to 6 questions
↳ You answer them
↳ Now the output actually has good context
Here's why most prompts fail:
"Write me a LinkedIn post about AI tools."
"Make it engaging."
"Keep it under 200 words."
One sentence.
No role.
No context.
No constraints.
That's like handing someone a blank page and saying "make it good."
The fix isn't a longer prompt.
It's a smarter setup.
Bad prompt: vague sentence, generic slop.
Okay prompt: clear task plus "ask me questions first."
Great prompt: .md files plus role plus task plus constraints plus "ask me questions first."
That last one is the difference between AI you delete and AI you ship.
📌 Want my AI playbooks? Join my free community (Link in bio)
♻️ Repost to help one person stop Googling their AI.
➕ Follow Ricky Waters for more on AI and LinkedIn growth.