Dear Reader,
Most prompts fail before you hit enter.
Not because of the AI model nor because of token limits or temperature settings. They fail because you haven’t actually figured out what you want.
I’ve been watching a pattern in my own work and in consulting. Someone sits down with ChatGPT or Claude, types out a request, gets something back, realizes it’s not quite right, refines the prompt, gets something closer, refines again, and twenty iterations later they’ve burned through context and patience for a result that’s “good enough.”
The problem isn’t prompt engineering. It’s that we’re generally asking AI to do our thinking for us — including the thinking about what we actually need.
The Vibe Prompting Trap
“Summarize this document.”
“Write me an email response.”
“Create a dashboard for my data.”
These prompts feel efficient. They’re short and can even get you to an output fast. But they’re also why you end up in correction loops.
Here’s what happens: You have a vague sense of what you want. AI interprets your vague request using its own assumptions. The output doesn’t match your mental image — because AI doesn’t have access to your mental image. So you correct. And correct. And correct.
Each correction is you discovering what you actually wanted in the first place.
The meta-prompting practitioners figured this out. They stopped telling AI what to do and started asking themselves what they needed first and the result is dramatically better outputs with fewer iterations.
But they didn’t invent a new technique. They rediscovered something systematic.
The Four Questions You’re Skipping
Every prompt that fails does so because you skipped at least one of these questions:
1. Purpose: What specific outcome do I need?
Not “a summary” — but what will you do with this summary? Is it for your own understanding? For a decision-maker who needs to approve something? For documentation?
“Summarize this research” produces generic output.
“Create a 300-word executive summary focused on ROI metrics and risk factors, because my CFO needs to decide on Q4 investment by Friday” produces useful output.
The difference isn’t prompt length. It’s purpose clarity.
2. Audience: Who receives this?
A technical team needs precision and specificity. Executives need conciseness and business framing. Customers need empathy and clarity. Internal teams need step-by-step actionability.
When you don’t specify your intended audience, AI will default to generic (and generic serves no one well).
3. Scope: What’s included—and what’s not?
Word count. Information sources. Level of detail. What you explicitly don’t want included.
“Give me an analysis” is unbounded. AI doesn’t know if you want a paragraph or a dissertation, surface observations or deep investigation, everything or just the relevant parts.
Scope constraints aren’t limitations. They’re clarity.
4. Tone: How should this feel?
Formal or conversational? Technical or accessible? Your voice or standard business? Cautious or confident?
AI will match what you ask for. If you don’t ask, you get its default — which is often generic corporate, thanks to the training data.
Why This Works
These four questions aren’t arbitrary. They force you to do the thinking that AI can’t do for you.
When someone asks Claude to “build a dashboard” and gets a messy result, the problem isn’t Claude’s capability. It’s that “build a dashboard” contains dozens of unstated assumptions about what kind of dashboard, what data, what format, who uses it, and what decisions it supports.
Meta-prompting — asking how you should ask — works because it surfaces these assumptions before execution instead of during correction loops.
The practitioners who’ve figured this out describe it as “asking AI how it would like to be asked.” But that framing buries the insight. What they’re really doing is systematically defining their own requirements before making a request.
Purpose. Audience. Scope. Tone.
If those four words sound familiar, it’s because they’re the PAST Framework — the same thinking structure that works for organizational AI strategy, team workflow design, and individual productivity.
The reason it works at every level is that the questions never change. Whether you’re defining a company-wide AI implementation or writing a single prompt, you still need to answer: What outcome? For whom? Within what boundaries? In what style?
From Theory to Practice
Here’s what this looks like applied:
Vibe prompt: “Analyze this customer feedback data”
PAST-structured prompt: “Analyze this customer feedback data to identify the top 3 recurring complaints and their root causes (Purpose: inform product roadmap decisions). Write for a product manager who needs to present recommendations to engineering (Audience: PM, not technical deep-dive). Focus only on complaints mentioned 5+ times; exclude one-off issues (Scope). Use clear problem statements with evidence counts, matter-of-fact tone (Tone).”
The second prompt takes 30 seconds longer to write. It produces dramatically better results on the first pass.
The math is simple: Would you rather spend 30 seconds thinking upfront, or 20 minutes in correction loops?
A practical note: If you’re unsure whether your prompt is clear enough, ask the AI itself. Before submitting your actual request, you can say: “I’m about to ask you to [describe task]. Can you help me refine this prompt using Purpose, Audience, Scope, and Tone?” The model will walk you through the questions and help you build a better request. You’re still doing the thinking—the AI is just making sure you’ve covered everything.
When to Apply This
Not every prompt needs this treatment.
If you’re asking what time it is in Tokyo, just ask. If you’re changing a background color from yellow to red, just say so.
But anything involving multiple steps, multiple possible interpretations, or outputs that matter—this is where the four questions pay off.
Complex analysis. Content creation. Process design. Anything where “good enough” isn’t actually good enough.
The rule of thumb: If you could imagine the output going three different directions based on how AI interprets your request, you haven’t been clear enough about Purpose, Audience, Scope, and Tone.
The Real Insight
Meta-prompting isn’t a technique. It’s a symptom of the same problem that shows up everywhere in AI implementation: people optimizing for speed instead of clarity.
The fastest prompt is rarely the most effective. The most sophisticated model doesn’t fix unclear thinking. The newest feature doesn’t compensate for not knowing what you want.
This is integration over capability again. A mediocre model with clear requirements produces better results than a powerful model with vague requests. The capability exists in both cases. The difference is in how systematically you apply it.
Framework thinking beats prompt tricks every time.
Your Challenge
Before your next substantial AI interaction, stop. Don’t type the prompt yet.
Answer these four questions first:
- Purpose: What specific outcome do I need, and what will I do with it?
- Audience: Who will use or read this output, and what do they need?
- Scope: What’s included, what’s excluded, and what constraints exist?
- Tone: What voice, format, and style serve the purpose and audience?
Write out your answers. Then write the prompt.
I’m betting you’ll get a better result in one pass than you usually get in five.
Let me know how it goes.
Until next week,
Jim
