Hi there. |
🎯 Just Joining Us? |
If this is your first email and you haven't followed from Day 1: Welcome! You're jumping into Day 4 of our 7-day JSON prompting course. To get up to speed and understand everything we've been building, I need to send you what we've learned from Day 1. |
Here's what to do: Reply to this email and let me know you're just joining us. In my response, I'll send you a complete summary of Days 1-3 so you have the full foundation. Then you can jump into Day 4 with the rest of the group. |
This takes 2 minutes to read and gives you everything you need to understand what's happening today. Don't skip this—the foundation matters. |
|
🎯 Where We Are (For Those Following From Day 1) |
Day 1: You learned JSON basics. |
Day 2: You used working prompts. |
Day 3: You built your own prompts from scratch. |
Day 4 (today): You make them better. |
Here's the honest truth: Your prompts won't be perfect on the first try. And that's actually fine. Professional prompt engineers don't build perfect prompts once—they build good prompts and then iterate them into great prompts. Today, you learn the iteration loop. |
|
|
📚 What You're Learning Today |
Four essential skills: |
How to recognize when a prompt needs work. What to look for in the output. How to diagnose what went wrong. The thinking process to identify the problem. How to fix it without rewriting everything. Surgical adjustments, not overhauls. The iteration loop that gets you to perfect. Test → Diagnose → Fix → Test again.
|
By the end of today, you'll be able to take any prompt that's "close but not quite" and make it exactly what you need. |
|
Part 1: Recognizing When A Prompt Needs Work |
🔍 What Does "Bad Output" Actually Look Like? |
Here's the thing: AI output isn't just "good" or "bad." It's usually in between. It's 60% perfect. Or 80% perfect. And your job is recognizing where it falls short. |
Real example: You send your blog post prompt to Claude. You get back a blog post that's... okay. But it's too formal. You wanted conversational. The examples aren't specific enough. The conclusion doesn't have a CTA. That's not a "bad prompt." That's a prompt that needs tweaking in three specific places. |
|
|
📋 The Five Types Of Output Problems (And What They Mean) |
Problem #1: The Tone Is Wrong |
What it looks like: Output is too formal when you wanted casual. Or too casual when you wanted professional. The voice doesn't match what you asked for. |
What went wrong: Your voice_requirements weren't clear enough. Or they weren't specific enough. |
Quick fix: Make your tone description more specific. |
Instead of "conversational," try "conversational—like texting a friend, not a corporate email." |
Problem #2: Missing Information |
What it looks like: Output is complete but missing something. No examples. No CTA. No intro. Something that was supposed to be there isn't. |
What went wrong: Your "must_include" section wasn't specific enough. |
Or it was there but worded vaguely. |
Quick fix: Add it to "must_include" more explicitly. Instead of "examples," write "3 specific, real-world examples." |
Problem #3: Too Long Or Too Short |
What it looks like: Output is 500 words when you wanted 1500. |
Or it's 3000 words when you wanted 800. Length is off. |
What went wrong: You either didn't specify word count or specified it vaguely. |
Quick fix: Add or clarify word count. |
Instead of "around 1500 words," write "exactly 1500-1800 words." |
Problem #4: Wrong Format Or Organization |
What it looks like: Output is one big paragraph when you wanted sections. |
Or it has headings when you didn't want them. Structure is off. |
What went wrong: Your output_format requirements were missing or unclear. |
Quick fix: Be very specific about format. Instead of "organized," |
write "organized with 3 clear sections, each with a heading, subheading, and 2-3 bullet points." |
Problem #5: Wrong Audience Or Level |
What it looks like: Output is too technical for your audience. Or too simplistic. |
Doesn't match who it's supposed to be for. |
What went wrong: Your audience description wasn't specific enough. AI doesn't understand your audience well. |
Quick fix: Be much more specific about your audience. |
Instead of "business owners," write "new entrepreneurs with technical backgrounds building their first startups." |
|
🎯 The Diagnostic Checklist: What To Look For |
When AI returns output, run through this checklist: |
✓ Does the tone match what I asked for? |
Read the first paragraph. Does it sound like what you specified? Yes = pass. No = fix tone. |
✓ Is all required information there? |
Check your "must_include" list. Is everything present? Missing something = add to requirements. |
✓ Is the length in the right ballpark? |
Quick word count. Too long or short = clarify word count specification. |
✓ Is the format right? |
Does it have the sections/headings/bullets you wanted? No = clarify output format. |
✓ Does it match the audience level? |
Would your target audience understand this? Is it too technical or too simple? No = refine audience description. |
|
Part 2: The Diagnostic Process (How To Know What To Fix) |
🔧 The Three-Question Diagnosis |
When output isn't quite right, ask yourself these three questions in order: |
Question 1: What specifically is wrong? |
Don't say "it's not good." Be specific. "The tone is too formal" or "it's missing examples" or "it's only 800 words but I needed 1500." Pinpoint the issue. |
Question 2: Which part of my prompt caused this? |
Look back at your prompt. Did you specify this? If you asked for "examples," did you say how many? Did you say how specific? If you didn't—that's the problem. |
Question 3: What's the minimum change I need to make? |
Don't rewrite your whole prompt. Just fix the broken part. If tone is wrong, fix the voice_requirements. If examples are missing, add to must_include. Surgical fix, not overhaul. |
|
📝 Real Example: Diagnosing A Blog Post Prompt |
You test your blog post prompt and get back output that's: |
Too formal (sounds like a textbook) Perfect length (1800 words) Has all required sections Good format with headers But... no CTA at the end
|
Running the diagnosis: |
What's wrong? |
Two things: (1) Tone is too formal, and (2) Missing CTA at the end. |
Which parts of my prompt caused this? |
Looking back: (1) My voice_requirements said "conversational" but wasn't specific enough. I didn't say "NO corporate language." (2) I didn't explicitly say "include CTA" in the requirements section. |
What's the minimum fix? |
Don't rewrite. Just: (1) Change voice_requirements to be more specific about avoiding formal tone. (2) Add "Include strong CTA at the end" to must_include. |
Part 3: Making The Fix (Surgical Changes) |
🔨 The Four Types Of Fixes (Most Common) |
FIX #1: Clarify Voice Requirements |
Your original: |
"tone": "conversational", "voice_requirements":
{ "style": "Write like you're talking to a friend",
"avoid": ["jargon"],
"include": ["examples"]
}
|
Your fix (add specifics to "avoid"): |
"tone": "conversational",
"voice_requirements": {
"style": "Write like you're texting a friend about this topic—casual, direct, no formality",
"avoid": ["corporate language", "formal tone", "complex sentences", "technical jargon"],
"include": ["real examples",
"step-by-step instructions", "casual language"]
}
|
FIX #2: Add Missing Requirements |
Your original: |
"requirements": { "must_include": [ "Introduction", "3 main sections", "Conclusion" ]
}
|
Your fix (add the missing CTA): |
"requirements": { "must_include": [ "Introduction that hooks readers", "3 main sections with real examples",
"Conclusion with strong CTA",
"At the end: a specific call to action telling readers what to do next" ] }
|
FIX #3: Clarify Output Format |
Your original: |
"output_format": { "include_headings": true, "include_bullets": true }
|
Your fix (be specific about how many and where): |
"output_format": { "include_headings": true,
"main_heading": "1 main title at the top", "subheadings": "1 subheading before each main section",
"bullet_points": "2-3 bullet points per section", "include_examples": "real, specific examples in each section"
}
|
FIX #4: Refine Audience Description |
Your original: |
"target_audience": "Small business owners"
|
Your fix (be specific about their knowledge and needs): |
"target_audience": "Small business owners who manage their own marketing",
"audience_details": { "who_they_are": "Solo entrepreneurs or small teams with 1-10 people",
"their_knowledge_level": "Non-technical, but comfortable with basic tools",
"what_they_want": "Practical, actionable tips they can implement in under an hour",
"their_frustration": "They don't have time for complex solutions"
}
|
|
Part 4: The Iteration Loop (The Complete Process) |
🔄 The Four-Step Loop That Gets You To Perfect |
Step 1: TEST |
Paste your prompt into ChatGPT or Claude. Get the output. Read it all the way through. |
Step 2: DIAGNOSE |
Ask the three diagnostic questions: (1) What's specifically wrong? (2) Which part of my prompt caused it? (3) What's the minimum fix? |
Step 3: FIX |
Make surgical changes. Don't rewrite. Just fix the broken part using one of the four fix types. |
Step 4: TEST AGAIN |
Paste the updated prompt in. See if it's better. If it is—great. If there's still an issue—go back to Step 2. |
Keep looping until:The output is 95%+ of what you wanted. Perfect is rare. Good enough to use without editing is the goal. |
📊 Real-World Example: Three Iterations To Perfect |
Iteration 1: First attempt |
You test. Output is 60% there. Tone is too formal. Missing specific examples. Missing CTA. |
Diagnosis: Voice requirements too vague. Requirements incomplete. |
Fix: Make voice_requirements more specific. Add examples and CTA to must_include. |
Iteration 2: Second attempt |
You test the updated prompt. Output is 85% there. Tone is better. Has examples now. Has CTA. But examples aren't specific enough—they're still too generic. |
Diagnosis: The "must_include" says "examples" but doesn't explain they need to be specific and real. |
Fix: Change "include real examples" to "include 3 specific, real-world examples that show exactly how this works in practice." |
Iteration 3: Third attempt |
You test the updated prompt. Output is 95% perfect. Tone is right. Examples are specific. CTA is there. Everything matches what you wanted. |
Done. Your prompt is ready to use repeatedly. |
|
⚡ Why Iteration Matters |
Iteration is how professional prompts are built. |
Nobody gets it perfect first try. What matters is knowing how to improve it quickly. |
After three iterations, you have a prompt that works. And every time you use it (on similar tasks), it works the same way. That's the power. |
Part 5: Common Mistakes & Quick Fixes |
❌ The Most Common Mistakes (And How To Fix Them) |
Mistake #1: Not Being Specific Enough |
What happens: You say "write a post about AI" and AI writes 500 words about AI in general. Too broad. |
The fix: Be much more specific. "Write a post about how AI tools help small business owners save time on email marketing specifically." |
Mistake #2: Conflicting Requirements |
What happens: You ask for "casual tone" but also "professional voice." AI gets confused and splits the difference awkwardly. |
The fix: Make sure your tone and voice align. Clarify: "Casual and friendly, but still professional—like talking to a trusted colleague." |
Mistake #3: Not Describing Your Audience Well |
What happens: You say "business owners" and AI writes for CEOs when you meant solopreneurs. |
The fix: Be detailed: "Solopreneurs with technical backgrounds building their first startups—they know code, but they're new to business." |
Mistake #4: Forgetting To Specify Word Count |
What happens: You get 800 words when you needed 2000. Or vice versa. |
The fix: Always include word_count. Not "around 1500" but "1500-1800 words exactly." |
Mistake #5: Vague "Must Include" Sections |
What happens: You ask to "include examples" and get 1-2 generic ones instead of 5 specific ones. |
The fix: Be explicit: "Include 5 specific, real-world examples that show exactly how this works." |
|
🚀 Your Challenge Today |
Here's what to do: |
Take the prompt you built on Day 3. (If you haven't, build one now using the Day 3 framework.) Test it. Paste it into ChatGPT or Claude. Get the output. Diagnose it. Use the three diagnostic questions. What's wrong? Which part caused it? What's the minimum fix? Make the fix. Use one of the four fix types. Don't rewrite. Just fix the broken part. Test again. See if it's better. Iterate until it's 90%+ perfect. Stop when you'd use it as-is without major editing. Reply and tell me: What did you fix? How many iterations did it take? Did your output get better?
|
Important: This isn't about perfection. It's about good enough to use. Usually 2-3 iterations gets you there. If it takes more—your initial prompt might need a bigger change (go back to Day 3 framework). |
|
⚡ Why This Matters |
Days 1-4: You've learned to build prompts and make them better. |
Days 5-7: You'll learn to combine them, organize them, and make them your actual workflow. |
By Day 7, this iteration skill is what keeps your entire system sharp. |
👀 What's Coming Tomorrow |
Day 5: Layering and combining techniques. You'll learn how to build prompts for complex, multi-step projects. How to do batch processing. How nesting becomes your secret weapon. |
But first—iterate today. Get your Day 3 prompt to perfect. |
Reply and show me the before and after. |
|
✍️ Show Me Your Iteration |
Reply with: |
Your original prompt: Paste the Day 3 prompt you built. |
What was wrong with the first output? Be specific. Tone? Missing info? Length? Format? |
What did you change? Which part of your prompt did you fix and how? |
How many iterations did it take? First try perfect? Took three tries? |
Your final prompt: Paste what you ended up with after iterations. |
Just joining us from Day 1? Reply and let me know, so I can send you the complete foundation from Days 1-3. |
Show me the improvement. This is where prompts become actually useful. |
Your Day 3 prompt was good. |
Today you're making it great. |
This is the difference between "knowing how to build prompts" and "actually using prompts every day." |
The iteration loop is where the magic happens—where your prompt transforms from "close enough" to "exactly what I need." |
Go test. Diagnose. Fix. Iterate. Repeat until it's perfect. |
— Prompt Guy |
P.S. The four-step iteration loop (Test → Diagnose → Fix → Test Again) is how every professional prompt engineer works. Do this with every prompt you build. By day 7, it'll be second nature. |