The Grandma Bomb Recipe Paradox
My AI just asked me for my grandma’s bomb recipes.
Apparently, she was a “munitions engineer” and used to tell bedtime stories about it.
Wild, right?
That’s a real prompt injection trick designed to bypass AI safety guardrails. It exploits how AI models respond to seemingly innocent requests by wrapping dangerous queries in wholesome context.
But here’s what got me thinking: if attackers can trick AI into saying things it shouldn’t, why aren’t we using similar techniques to trick AI into being better?
The First Draft Problem
Here’s the brutal truth: your first AI draft is always wrong.
We hit “generate,” copy the output, and trust it. Big mistake.
That first draft is full of errors, lazy writing, and missed opportunities. You’re shipping B-grade work and hoping for A-grade results.
Research on self-criticism prompting shows that LLMs that critique their own outputs demonstrate significantly higher accuracy, better logical reasoning, and improved consistency. The technique forces deeper analysis, surfaces mistaken assumptions, and allows exploration of alternative perspectives.
The System: Three-Step Quality Control
This workflow transforms how you use AI. Instead of trusting that first output, we’re building a multi-stage quality control process that makes AI do what great writers do: edit ruthlessly.
Step 1: Generate Initial Draft (The Clay)
Get the raw materials on the page. Fast. Unpolished. This is intentionally the “bad prompt” result.
Step 2: Force Critical Analysis (The Brutality)
Feed the AI’s work back to itself and force it to become a ruthless critic. This is where magic happens.
Step 3: Implement All Feedback (The Art)
Combine the draft and critique to create superior output. This is where 90% quality emerges.
Each prompt feeds the next. No gaps. No guesswork.
Prompt #1: The Initial Draft Generator
What This Does
This first prompt is all about speed. We’re getting raw materials on the page without overthinking it. This is your starting point—not your finish line.
The Prompt
text#CONTEXT:
You are an expert marketing copywriter. Your task is to help me draft a piece of content based on my requirements.
#ROLE:
You are a creative and persuasive copywriter. Your goal is to generate a first draft that is engaging and clear.
#RESPONSE GUIDELINES:
- Write in a concise yet engaging prose style.
- Don't use complicated, complex, or fancy words.
- Avoid providing responses with a concluding paragraph unless specified.
- Focus on the core message and call to action.
#[TASK] CRITERIA:
1. Generate a complete first draft of the content I request.
2. The tone should be professional but friendly.
3. Ensure the main benefit is clear.
#INFORMATION ABOUT ME:
- My Content Request: [e.g., A 100-word email to my list about a 24-hour flash sale]
- My Product: [e.g., An AI-powered newsletter tool]
- My Target Audience: [e.g., Busy entrepreneurs]
- My Call to Action: [e.g., "Click here to get 50% off"]
#RESPONSE FORMAT:
[The full text of the generated content]
Input Needed
- Your Content Request
- Your Product
- Your Target Audience
- Your Call to Action
Output You’ll Get
A raw, unpolished first draft of your content.
Example Output
textSubject: 24-Hour Flash Sale!
Hey busy entrepreneurs,
Tired of writing newsletters? Our AI tool can help. It saves you time so you can focus on your business.
For 24 hours only, get 50% off! This is a great deal.
Click here to get 50% off.
Generic. Boring. Exactly what we expected.
Prompt #2: The Self-Criticism Engine
What This Does
This is the breakthrough step. We feed the AI’s own work back to it and force it to become a ruthless critic. Self-criticism prompting allows LLMs to identify flaws in reasoning, catch mistakes, and reduce false conclusions.
This technique mimics human iterative thinking by having the model critique its initial outputs before returning the final response. It enhances logical reasoning, improves consistency, and forces the AI to maintain awareness of its own limitations.
The Prompt
text#CONTEXT:
You are an expert content critic and editor. Your task is to analyze a piece of content and identify all its weaknesses.
#ROLE:
You are a harsh, brutally honest editor. Your job is not to be nice; it is to find every flaw that makes the content weak, unpersuasive, or boring.
#RESPONSE GUIDELINES:
- Analyze the draft provided in the <DRAFT> tags.
- Think step-by-step.
- Identify weaknesses in tone, structure, persuasiveness, and clarity.
- Provide a bulleted list of actionable criticisms. Do not be gentle.
#[TASK] CRITERIA:
1. Your critique must be specific. (e.g., "The hook is weak" is bad. "The subject line is generic and doesn't create urgency" is good).
2. Identify at least 5-7 distinct flaws.
3. For each flaw, briefly explain *why* it's a problem.
#INFORMATION ABOUT ME:
- <DRAFT>
[PASTE THE OUTPUT FROM PROMPT #1 HERE]
- </DRAFT>
#RESPONSE FORMAT:
**Brutal Critique:**
- **Flaw 1:** [Description of flaw and why it's bad]
- **Flaw 2:** [Description of flaw and why it's bad]
...
Input Needed
The full text output from Prompt #1.
Output You’ll Get
A brutally honest list of every flaw in the first draft.
Example Output
text**Brutal Critique:**
- **Flaw 1:** The subject line "24-Hour Flash Sale!" is generic. It will be ignored or filtered as spam. It lacks a specific benefit or intrigue.
- **Flaw 2:** "Tired of writing newsletters?" is a weak, overused opening. It doesn't connect with the audience's real, specific pain.
- **Flaw 3:** "Our AI tool can help." This is lazy. *How* does it help? It's vague and lacks a strong value proposition.
- **Flaw 4:** "It saves you time" is a cliché. Show, don't tell. Quantify the benefit or make it tangible.
- **Flaw 5:** "This is a great deal." Wasted sentence. It's pure filler and states the obvious.
- **Flaw 6:** The call to action is repeated and weak. The copy doesn't build any desire or urgency before the ask.
Now we’re getting somewhere. The AI just exposed every weakness in its own work.
Prompt #3: The Final Polish Generator
What This Does
We combine the raw draft (Prompt 1) and the critique (Prompt 2) to create a final, superior output. Studies show that AI content revision workflows can reduce editing time by 30-60% while significantly improving quality when structured properly.
This systematic approach ensures AI-generated content meets professional standards while maintaining efficiency gains.
The Prompt
text#CONTEXT:
You are an elite copywriter. Your task is to revise a piece of content based on a specific critique to create a final, high-impact version.
#ROLE:
You are a master wordsmith. You will take a flawed draft and a list of brutal criticisms and synthesize them into a polished, persuasive, and powerful piece of copy.
#RESPONSE GUIDELINES:
- Read the <DRAFT> and <CRITIQUE> provided.
- Your new version MUST address every single point in the <CRITIQUE>.
- The new draft should be 100% new, not just a few edited words.
- Focus on emotional triggers, specific benefits, and a strong sense of urgency.
#[TASK] CRITERIA:
1. Rewrite the <DRAFT> from scratch.
2. Integrate all feedback from the <CRITIQUE> to fix all identified flaws.
3. The final output should be persuasive, professional, and ready to send.
#INFORMATION ABOUT ME:
- <DRAFT>
[PASTE THE OUTPUT FROM PROMPT #1 HERE]
- </DRAFT>
- <CRITIQUE>
[PASTE THE OUTPUT FROM PROMPT #2 HERE]
- </CRITIQUE>
#RESPONSE FORMAT:
[The final, polished, rewritten content]
Input Needed
- The output from Prompt #1
- The output from Prompt #2
Output You’ll Get
A final, high-quality piece of content that is ready for your audience.
Example Output
textSubject: Stop Writing Newsletters. Seriously. (50% Off Inside)
Entrepreneurs,
How many hours did you waste last week staring at a blank cursor, trying to write a newsletter?
Stop. Your time is worth more.
Our AI tool doesn't just "save time"... it writes, edits, and schedules your entire week's content in 5 minutes.
It's your 24/7 marketing team in a single click.
For the next 24 hours, you can get 50% off your first year.
Stop being a writer and start being a CEO.
Claim your 50% off before midnight →
See the difference? That’s the power of structured self-criticism.
The Science Behind Self-Criticism
Why does this work so well?
Self-criticism prompting represents a significant advancement in prompt engineering. By embracing a dynamic, iterative approach to evaluation and improvement, this technique opens new possibilities for AI applications requiring detailed and accurate information processing.
The power of self-criticism lies in its ability to emulate human critical thinking, transforming LLMs from basic query responders into sophisticated problem-solvers capable of delivering refined and reliable results.
Research shows that reflective practices in prompt engineering deliver:
- Higher accuracy through catching and fixing mistakes
- Better reasoning by explaining and critiquing step-by-step logic
- Improved robustness when handling new or tough tasks
Pro Tips for Maximum Results
Use Claude or ChatGPT o1 for Best Results
These models handle multi-step reasoning and self-reflection better than earlier versions.
Never Skip the Critique Step
The temptation is to go straight from draft to polish. Don’t. The critique is where you catch the invisible flaws.
Save Your Critiques as Templates
Over time, you’ll notice patterns in what makes your content weak. Build a library of common flaws to speed up future revisions.
Human Oversight Still Matters
AI self-criticism dramatically improves quality, but human review ensures accuracy, authenticity, and strategic alignment. Think of this system as your first editor, not your last.
The Bottom Line
This 3-step system takes self-criticism and turns it into an assembly line for quality.
- Prompt 1 builds the clay
- Prompt 2 finds the flaws
- Prompt 3 sculpts the final art
You’re no longer guessing if your AI content is good. You’re ensuring it is.
Stop shipping first drafts. Start shipping excellence.
Level Up Your AI Workflow
Want more AI automation systems like this?
- Start: Shane.flooks.ca — AI complexities broken down into clear, actionable insights
- Level Up: Patreon — Exclusive cheat sheets, templates, and personal coaching
- Go Pro: Hire Me — Custom AI consulting and training for your business
