🔑 This prompt method changes everything

In partnership with

Reading time: 7 minutes

Greetings from above,

Why did the AI go back to school? Because it wanted to Self-Refine its education!

Last week, I was struggling with generic AI responses for a client project. The outputs were technically correct but lacked that special touch.

Then I discovered Self-Refine - within hours, my client was praising the "remarkable improvement" in our AI-written content.

Today, we'll talk about:

• What Self-Refine is and why it matters

• How to implement this technique with any AI model

• Real-world examples that demonstrate the power of Self-Refine

• A step-by-step framework you can use immediately

Let's dive in!

Prompt, run, edit, and deploy full-stack web and mobile apps

Say hello to Bolt.new, your AI-powered web development agent that brings full stack coding to your browser with no setup required. Built on StackBlitz’s WebContainers, Bolt lets you build, edit, and test web applications in real time with simple, chat-based prompts. Just tell the agent what you want to build or change, and watch your code update instantly.

Bolt supports today’s most popular web languages and frameworks, so you can get started fast whether you’re building a React app, deploying with Netlify, or integrating a backend with Supabase for auth, storage, and more. Need a mobile version too? Bolt works with Expo to streamline mobile app development.

Whether you’re a solo builder or part of a product team, Bolt is your fast lane to production-ready code right from your browser.

Try it today

❓ THIS WEEK'S READER QUESTION

We've all been there, Marcus. You ask ChatGPT for something, and while the answer isn't wrong, it lacks that special sauce.

It's like ordering a pizza and getting one with all the right toppings but somehow still tastes bland.

What if I told you there's a way to make the AI critique and improve its own work?

No complex prompting required.

Just a simple technique that leverages what the AI already knows.

THE SELF-REFINE TECHNIQUE: MAKE AI FIX ITS OWN ANSWERS

Prompt engineering isn't always about writing better inputs.

Sometimes it's about creating a process that allows the AI to improve itself.

That's exactly what Self-Refine does.

WHAT IS SELF-REFINE?

Self-Refine is a prompting technique where a single AI model plays three distinct roles:

1️⃣ Generator: Creates an initial response to your query

2️⃣ Critic: Reviews and critiques that response based on specific criteria

3️⃣ Improver: Enhances the original response using the feedback

This cycle can repeat until the output reaches your desired quality level.

The best part? This technique works with any AI model - even GPT-4.

And it delivers an average performance boost of ~20% across diverse tasks.

That's like getting a free upgrade without changing your subscription!

HOW SELF-REFINE WORKS

Think of Self-Refine as having an AI coach for your AI.

The same model iterates to improve its own writing, code, or messaging—without any additional training, extra data, or tools.

It's like having a writing workshop inside a single prompt.

The key is in how you structure the three-step process:

Key benefits:

• Improves output quality by approximately 20%

• Works with any AI model (including GPT-4)

• Requires no special tools or technical skills

• Creates more relevant, specific, and persuasive content

• Addresses common AI pitfalls like generic responses

REAL-WORLD EXAMPLES OF SELF-REFINE

Let's see how this technique transforms outputs in different scenarios.

1. User Inquiry Response (Table Tennis)

This example shows how Self-Refine can transform a generic response into something more engaging and personalized.

Why this matters:

• Improves customer satisfaction by providing more helpful responses

• Creates more personalized interactions that build trust

• Reduces the need for follow-up questions

STEP 1 - INITIAL PROMPT:
I need you to respond to this user message in a helpful way:

User: "I am interested in playing table tennis."

What it does: Generates a basic response that addresses the query but remains generic.

Initial output: 

Perfect for:

• Customer service teams handling inquiries

• Community managers responding to member questions

• Support agents who need to provide helpful information quickly

Implementation tip: Start with a simple, straightforward request to establish a baseline response.

STEP 2 - FEEDBACK PROMPT:
Review the following response to a user message. Provide specific, actionable feedback on:

1/ Engagement: How well does it address the user's needs?
2/ User understanding: Does it demonstrate understanding of the user's situation?

User: "I am interested in playing table tennis."

Response:
"I'm sure it's a great way to socialize, stay active and have fun. If you're looking to get started, you might want to find a local club or community center that offers tables and perhaps lessons for beginners."

What it does: Forces the AI to critically evaluate its own response against specific criteria.

Feedback output: 

STEP 3 - REFINE PROMPT:
Improve the following response based on the feedback provided:

User: "I am interested in playing table tennis."

Original response:
"I'm sure it's a great way to socialize..."

Feedback:
"Too generic, lacks specificity or deeper understanding."

What it does: Instructs the AI to create an improved version based on its own critique.

Refined output: 

2. Business Use Case - Cold Email

This example demonstrates how Self-Refine can transform a weak marketing message into a compelling outreach email.

Why this matters:

• Increases response rates by creating more targeted messaging

• Demonstrates understanding of specific customer pain points

• Creates a stronger call-to-action that drives conversions

STEP 1 - INITIAL PROMPT:
Write a cold outreach email offering SEO services to early-stage eCommerce startups.

What it does: Generates a basic email draft for the specified purpose.

Initial output: 

Perfect for:

• Marketing teams crafting outreach campaigns

• Sales professionals looking to improve their pitch emails

• Founders reaching out to potential clients or partners

Implementation tip: Be specific about your target audience and purpose in the initial prompt.

STEP 2 - FEEDBACK PROMPT:
Review this cold email draft.

Provide detailed feedback on:
- Clarity and value proposition
- Relevance to target (early-stage eCommerce startups)
- Tone and call-to-action strength

Email:
"Hi, we provide SEO services that help startups grow. Let's chat!"

What it does: Evaluates the email against specific marketing criteria.

Feedback output: 

STEP 3 - REFINE PROMPT:
Rewrite the email using the feedback above. Make it more relevant, specific, and persuasive.

What it does: Creates an improved version that addresses all the identified weaknesses.

Refined output: 

HOW TO IMPLEMENT SELF-REFINE IN YOUR WORKFLOW

Creating a Self-Refine process is surprisingly simple. Here's how to do it:

1. Set up your three prompts:

• Generate Prompt: Ask the AI to create an initial response

• Feedback Prompt: Ask the AI to evaluate its response based on specific criteria

• Refine Prompt: Ask the AI to improve the response based on the feedback

2. Define clear evaluation criteria:

Be specific about what aspects you want the AI to critique:

• Relevance to the target audience

• Clarity and specificity

• Tone and engagement

• Action items and next steps

• Any industry-specific requirements

3. Iterate as needed:

For best results, run the feedback-refine loop 2-3 times.

Each iteration makes the output sharper and more effective.

Key implementation steps:

• Start with a clear, specific initial prompt

• Provide 2-4 explicit criteria for the feedback stage

• Ask for specific improvements in the refine stage

• Repeat until satisfied with the quality

DIVE DEEPER (ON MY TWITTER)

🏆 WEEKLY PROMPT CHALLENGE

This week, I challenge you to try this Self-Refine prompt:

I want you to act as a Self-Refine system with three components:

1. CREATOR: Generate an initial response to my query
2. CRITIC: Evaluate that response based on specificity, engagement, and actionability
3. IMPROVER: Create an enhanced version addressing the critique

My query is: [INSERT YOUR QUERY HERE]

Please show me all three steps: initial response, critique, and refined version.

Share your results with me by replying to this email!

The best submission gets featured next week.

💡 PROMPT TIP OF THE WEEK: SPECIFICITY IN FEEDBACK REQUESTS

When using Self-Refine, the quality of your feedback criteria makes all the difference.

Generic feedback leads to generic improvements. Specific feedback drives targeted enhancements.

Before:

Review this text and provide feedback.

After:

Review this text and provide specific feedback on:
1. How well it addresses the pain points of [specific audience]
2. The clarity of the main value proposition
3. The effectiveness of the call-to-action
4. The use of concrete examples vs. general statements

This approach gives the AI clear dimensions to evaluate, resulting in much more useful feedback.

It's like giving a chef specific flavor notes to enhance rather than just saying "make it taste better."

Try this approach with your next Self-Refine prompt and watch the quality of your outputs soar.

SUMMARY

• Self-Refine is a three-step process where an AI generates, critiques, and improves its own work

• This technique delivers a ~20% performance boost across various tasks

• No special tools or models required - works with any AI including GPT-4

• Most effective when using specific evaluation criteria

• Can be applied to writing, coding, marketing, customer service, and more

📚FREE RESOURCES📚

WRAP UP: What you learned today:

1. The Self-Refine technique - How to make AI improve its own outputs through structured feedback

2. Implementation framework - A three-step process you can apply to any AI task

3. Real-world examples - How Self-Refine transforms generic responses in customer service and marketing

Remember, the best prompt engineers don't just write better prompts - they create systems that leverage the AI's existing capabilities in clever ways. Self-Refine is the perfect example of working smarter, not harder with AI.

And as always, thanks for being a part of my lovely community,

Keep learning,

🔑 Alex from God of Prompt

P.S. What's your biggest challenge with getting quality AI outputs? Reply to this email with your prompting problems, and I might feature a solution in next week's newsletter!