Picture this: you need to write a comprehensive marketing plan for a new product launch. You open ChatGPT and type:
"Write a marketing plan for my new fitness app."
The AI produces five pages of generic content that misses your target audience, ignores your budget constraints, and suggests tactics you can’t execute. You spent tokens and time on something you can’t use.
Now imagine a different approach.
First, you ask for an outline of key sections a marketing plan needs. You review it, adjust for your situation, and approve the structure. Then you tackle each section one by one, asking for research on your target market, then competitor analysis, then channel recommendations.
At each step, you review, refine, and approve before moving forward. The final plan is tailored, actionable, and actually useful.
This is the power of the Plan-Draft-Critique Method.
👋 Julley , I'm Dheeraj, an AI systems builder.
I build production-grade AI systems at work by day and ship my own products by night, 9 and counting, including SubflowAI and the Content OS Agents Toolkit. This newsletter is the bridge between those two worlds. Every system, every build, documented step by step.
Join 1,800+ builders getting the exact AI setups, prompts, and workflows that actually work in your business.
What is AI task decomposition?
AI task decomposition is the practice of breaking a complex task into focused phases instead of asking AI to do everything at once. The Plan-Draft-Critique Method structures this into three steps: first Plan (outline the approach), then Draft (create the content), then Critique (review and improve). Three focused prompts consistently beat one giant request.
Instead of asking AI to do everything at once and hoping for the best, you stage the work into planning, drafting, and critiquing phases. Each phase builds on the previous one, errors get caught early, and the final output is dramatically better.
In this lesson, you’ll learn:
what prompt chaining is and why it works,
how to decompose complex projects into manageable steps,
when to use list plans versus tree plans,
how to write critique prompts that catch errors, and
how to use debate prompts for exploring multiple perspectives.
These techniques transform AI from a guess-and-hope tool into a reliable workflow partner.
Prompt Engineering Course - Complete Series
Get Reliable JSON from LLMs: Structured Output Prompting Guide
AI Task Decomposition: The Plan-Draft-Critique Method ← You are here
Real World Analogy
Complex tasks fail when you ask AI to do everything at once. Break them into phases (plan, draft, critique), and suddenly AI becomes a reliable partner, outputs improve dramatically, and you stay in control at every step.
Imagine you hire an architect to design your dream house. You have two choices.
Option A: You tell the architect “Build me a house” and walk away.
Six months later, they present you with a finished house. It has rooms you don’t need, missing features you wanted, a layout that doesn’t match how you live, and it’s over budget. But it’s built. You’re stuck with expensive changes or accepting something that doesn’t work.
Option B: The architect first shows you three rough sketches of different layouts. You pick one and suggest changes. Then they create detailed floor plans for each room. You review and refine. Next comes material selections, then contractor bids, then construction begins with regular check-ins at each phase.
The final house matches your vision because you stayed involved at every decision point.
Ask AI to “write a report” and you get Option A - a complete output that might miss the mark entirely. Apply the Plan-Draft-Critique Method and you get Option B - a collaborative process where you guide the outcome at each step.
The best part? With AI, each stage takes seconds or minutes instead of weeks. You can iterate rapidly, catch problems early, and produce results that would take days in a single afternoon.
What Is Prompt Chaining (And Why You Need It)
Before we get into the Plan-Draft-Critique framework, it helps to understand the underlying concept: prompt chaining.
Prompt chaining means splitting a big task into smaller steps and sending each step to the AI separately. The output of one prompt becomes the input for the next. Instead of one giant prompt that overwhelms the model, you guide it through a process - plan first, write second, review third.
Here’s why AI models give better results with smaller, focused prompts: they have limited attention. When you give an AI a massive prompt with multiple objectives, it has to juggle everything at once. Quality drops on each part because the model is spreading its processing across too many goals simultaneously.
A focused prompt with one clear objective gets the model’s full “concentration.” The task is simpler. The constraints are clearer. The output is better.
That’s prompt chaining. And the Plan-Draft-Critique Method is the most practical version of it for creators and solopreneurs.
Learn with me: the Claude Code cohort
I run the Claude Code cohort on Maven: six live sessions across six weeks, building a real AI system with me. If you would rather build it alongside me than read about it, this is the room.
Each cohort is a small, hands-on group, so seats are limited.
Use code GENAI20 for 20% off. Check the Syllabus →
The Plan-Draft-Critique Method: Breaking Big Tasks Into AI-Sized Pieces
The Plan-Draft-Critique framework has three phases. Each phase uses a separate, focused prompt. Each phase builds on what came before.
Here’s the complete flow:
Plan - You describe the goal. The AI outlines the structure. You review and refine before anything gets written.
Draft - You tackle one section at a time, using the approved plan as a guide. AI writes; you review before moving to the next section.
Critique - You ask the AI to evaluate what’s been written against specific criteria. You get actionable feedback, not vague suggestions.
Step 1: Plan - Mapping Out the Task
The planning stage creates the structure before filling in details. This prevents wasted effort on content that doesn’t fit your needs.
List Plan (Sequential)
A simple numbered or bulleted list of items to address. Best for straightforward tasks with clear sequential steps.
Create an outline for a blog post about email productivity tips.
Output format:
1. Introduction
2. Main point 1
3. Main point 2
etc.Tree Plan (Hierarchical)
A nested structure showing relationships and dependencies. Best for complex projects where parts connect to each other.
Create a hierarchical plan for launching a new feature.
Output format:
1. Pre-Launch Phase
1.1 Market Research
1.1.1 User surveys
1.1.2 Competitor analysis
1.2 Technical Preparation
1.2.1 Infrastructure setup
1.2.2 Testing protocols
2. Launch Phase
etc.Planning Prompt Template
You are a strategic planner. Create a detailed plan for [task].
Goal: [what you want to achieve]
Constraints: [budget, time, resources]
Success criteria: [how to measure success]
Create a [list/tree] plan that includes:
- All major steps or sections
- Dependencies between items (if tree plan)
- Time estimates for each part
- Risks or challenges to consider
Do NOT write the content yet, only the plan.The key phrase: “Do NOT write the content yet, only the plan.” This keeps the model focused on structure instead of jumping ahead.
Step 2: Draft - Getting the First Version
Once you have an approved plan, tackle each section individually. This prevents context drift, maintains quality, and lets you course-correct as you go.
Drafting Prompt Template
You are a [role]. Write [specific section] according to this approved plan:
[paste relevant section from plan]
Context from previous sections:
[brief summary of what's been written so far]
Requirements:
- Target audience: [who]
- Tone: [professional/casual/technical]
- Length: [word count or paragraph count]
- Key points to include: [list]
Write only this section. Do not continue to other sections.Why this works:
Focused scope: Model handles one clear task at a time
Context continuity: Previous work informs current section
Quality control: You review each piece before moving forward
Easy iteration: Rewrite one section without touching the rest
Clear boundaries: “Write only this section” prevents overrun
Step 3: Critique - Using AI to Review Its Own Work
The critique stage catches errors, inconsistencies, and missed opportunities before you consider the work done. This is where the Plan-Draft-Critique Method really earns its name.
Critique Prompt Template
You are a critical reviewer. Evaluate this [content type] and identify specific improvements.
Content to review:
[paste content]
Evaluate against these criteria:
1. Clarity: Is everything easy to understand?
2. Completeness: Are any important points missing?
3. Accuracy: Are there any factual errors or unsupported claims?
4. Consistency: Do all parts fit together logically?
5. Audience fit: Does this match the target audience's needs?
For each issue found:
- Quote the specific part
- Explain the problem
- Suggest a concrete fix
Prioritize issues by impact (High/Medium/Low).What makes a good critique prompt:
Specific criteria: Not “make it better” but “check clarity, completeness, accuracy”
Concrete feedback: Quote exact problems, not vague comments
Actionable fixes: Suggest what to change, not just what’s wrong
Prioritization: Focus on high-impact issues first
This article shows you the method. But the first time you apply Plan-Draft-Critique to a real project, you will spend 30 to 60 minutes just getting your prompt chains right before useful output appears.
Inside PluggedIn, I keep the ready-made prompt chain templates for the most common project types so you can skip that setup friction entirely.
What Is Prompt Chaining in Practice: Real Plan-Draft-Critique Examples
Let’s see the Plan-Draft-Critique framework applied to real scenarios you might actually face.
Example 1: API Integration Guide
Single-prompt approach (often fails):
Write a technical guide for integrating our API.
Plan-Draft-Critique approach (reliable):
Stage 1 - Plan:
Create an outline for an API integration guide. Include sections for:
- Prerequisites
- Authentication setup
- Core endpoints
- Code examples
- Error handling
- Rate limits
Output: numbered list of sections with 2-3 subsections each.Stage 2 - Draft (one section):
Write the "Authentication Setup" section of the API guide.
Context: This follows the Prerequisites section where users already have an API key.
Requirements:
- Explain OAuth 2.0 flow
- Include code example in Python
- Cover common errors
- 200-300 words
Write only this section.Stage 3 - Critique:
Review the Authentication Setup section for:
1. Technical accuracy
2. Code example completeness
3. Error coverage
4. Beginner-friendliness
List specific improvements with line numbers.Example 2: Marketing Campaign Plan
Stage 1 - Tree Plan:
Create a tree plan for a product launch campaign.
Product: AI writing assistant for content creators
Budget: $10,000
Timeline: 8 weeks
Goal: 500 signups
Include:
- Pre-launch activities
- Launch week tactics
- Post-launch nurture
- Success metrics for each phase
Format as nested list showing dependencies.Stage 2 - Draft (one branch):
From the approved campaign plan, write the "Pre-launch: Influencer Outreach" section.
Plan context:
- Target: 10 micro-influencers in content creation space
- Budget allocated: $2,000
- Timeline: Weeks 1-4
Include:
- Influencer selection criteria
- Outreach email template
- Collaboration terms
- Tracking metrics
Length: 300-400 wordsStage 3 - Debate:
Debate two approaches for the influencer outreach:
Option A: Pay for reviews ($200 per influencer)
Option B: Free access + affiliate program (20% commission)
Context: Budget is $2,000, goal is authentic promotion.
Present arguments for both sides, then recommend an approach with reasoning.Example 3: Quarterly Business Review
Stage 1 - Plan:
Outline a quarterly business review for e-commerce store.
Data available:
- Revenue: $150K (up 12%)
- Orders: 890 (up 8%)
- Return rate: 15% (up 3%)
- Top products: shirts, shoes, accessories
Create section outline that tells a coherent story.Stage 2 - Draft sections in order:
Section 1: Executive Summary
Write 2-3 paragraphs hitting highlights and key concern (return rate).
150 words.
[Review and approve]
Section 2: Revenue Analysis
Break down revenue by product category. Explain 12% growth drivers.
250 words with bullet points for each category.
[Review and approve]
... continue for each sectionStage 3 - Final Critique:
Review the complete quarterly report for:
1. Data consistency across sections
2. Narrative flow between sections
3. Action items clarity
4. Executive summary accuracy
List any disconnects or unclear points.Debate Prompts: Exploring Multiple Perspectives
The Plan-Draft-Critique Method also includes a variation for decisions: debate prompts. These ask AI to argue multiple sides before recommending an approach. This catches blind spots and reveals tradeoffs.
Debate Prompt Template:
You are a strategic advisor. I need to decide: [decision to make]
Context: [relevant background]
Analyze this decision by presenting:
Side A: Arguments FOR [option A]
- 3 strongest benefits
- Supporting evidence or examples
- Potential risks and mitigations
Side B: Arguments FOR [option B]
- 3 strongest benefits
- Supporting evidence or examples
- Potential risks and mitigations
Comparison:
- Key tradeoffs between approaches
- What assumptions each side depends on
- What factors would tip the decision one way or other
Recommendation:
Based on the analysis, which approach do you recommend and why?When to use debate prompts:
Making strategic decisions with no obvious right answer
Evaluating different approaches to a problem
Challenging your own assumptions
Presenting options to stakeholders
Understanding tradeoffs before committing
The Plan-Draft-Critique Refinement Loop
For critical content, add a refinement loop after the critique:
Plan an outline structure
Draft with sections
Critique to find issues
Revise specific problems
Re-critique to verify fixes
Repeat until quality threshold met
Refinement Prompt:
Based on this critique:
[paste critique feedback]
Revise the content to address all High priority issues.
For each fix:
- State which issue you're addressing
- Show the revised text
- Explain why the revision solves the problem
Do not change parts that weren't critiqued.This creates a feedback loop that dramatically improves quality with minimal manual editing. The Plan-Draft-Critique cycle repeats until you’re satisfied, with each iteration focused on specific, identified problems rather than vague “make it better” requests.
When Should You Chain Your Prompts?
Not every task needs the full Plan-Draft-Critique treatment. Here’s a quick decision guide:
Use the Plan-Draft-Critique Method when:
The task has multiple distinct sections or phases
You need to review and approve before moving forward
The output feeds into another process or document
The task is longer than 500 words
Getting it wrong is expensive or time-consuming to fix
Quality matters more than speed
Single prompt is fine when:
The task is short and focused (under 200 words)
There’s only one logical output format
You’re generating options to choose from, not a final deliverable
Speed matters more than quality for this specific use
The signal: if you’d normally review a human’s draft before they continue working, you should use the Plan-Draft-Critique approach.
Why Breaking Tasks Into Stages Gives You Control
Breaking tasks into stages isn’t just about getting better outputs. It’s about:
Control: You guide the direction at every decision point, not just at the start.
Quality: Errors caught early are easier and cheaper to fix than rebuilding from scratch.
Learning: Seeing the planning and critique stages teaches you how to think about the problem better.
Reusability: Your planning prompts become templates you can reuse for similar projects.
Confidence: When you review work at each stage, you trust the final output because you shaped it.
The difference between asking AI to “write a report” and systematically planning, drafting, and critiquing is like the difference between ordering takeout and cooking a meal. Takeout is fast but might not match your taste. Cooking takes more steps but gives you exactly what you want.
Once you’ve built a Plan-Draft-Critique workflow for one type of task, it becomes a reusable system. Your “newsletter outline prompt” and “critique prompt for marketing copy” are now tools you run, not tasks you reinvent.
This is the bridge to production workflows, which we cover in ChatGPT and Claude: Build Production AI Workflows.
Key Takeaways
Break complex tasks into three stages: plan, draft, critique
Planning prevents wasted effort on content that doesn’t fit your needs
List plans work for sequential tasks, tree plans for complex projects with dependencies
Draft one section at a time to maintain quality and prevent context drift
Critique prompts should specify concrete criteria and request actionable fixes
Debate prompts reveal tradeoffs by arguing multiple perspectives before deciding
The phrase “Do NOT write content yet, only the plan” keeps AI focused on structure
Review and approve each stage before moving to the next
Iterative refinement loops (draft, critique, revise, repeat) produce highest quality
Multi-stage workflows give you control, improve quality, and build reusable templates
Practice Exercises
Exercise 1: Pick a real task you need to complete this week (email, report, presentation). Write a planning prompt that outlines the structure without drafting content. Test it and see if the plan makes sense.
Exercise 2: Take something you recently wrote (email, document, post). Write a critique prompt with 4-5 specific evaluation criteria. Run it and see what issues it finds.
Exercise 3: Think of a decision you’re facing with two valid options. Write a debate prompt that argues both sides thoroughly. Does the analysis reveal any tradeoffs you hadn’t considered?
Exercise 4: Take a complex task and break it into stages:
Write the planning prompt
Get and approve the plan
Draft the first section
Critique that section
Revise based on feedback
Compare the final quality to what you’d get from a single prompt.
Exercise 5: Choose a project with dependencies (like planning an event or launching a feature). Create a tree plan showing how tasks relate to each other. Does visualizing dependencies reveal any sequencing issues?
Success check: Your staged approach should produce better quality results than single-prompt attempts, and you should feel more confident in the outputs because you guided them at each step.
Frequently Asked Questions
What is the Plan-Draft-Critique method?
The Plan-Draft-Critique method is a three-step AI prompting framework where you first ask AI to plan the structure of a task, then draft it section by section, then critique the output against specific criteria. It produces dramatically better results than asking AI to complete complex tasks in one prompt, because each phase is focused and you stay in control at every step.
What is prompt chaining in AI?
Prompt chaining means splitting a big task into smaller steps and sending each step to the AI separately. The output of one prompt becomes the input for the next. For example: Prompt 1 generates an outline, Prompt 2 writes content based on that outline, Prompt 3 reviews the content. Each link in the chain is simple and focused, which leads to better overall quality than one complex prompt.
How do I break down complex tasks for AI?
Use the Plan-Draft-Critique method. First, give AI your goal and ask it to create a structure or outline (don’t let it write content yet). Review and adjust the outline. Then write one section at a time using the approved outline as context. Finally, ask AI to critique what’s been written against specific criteria. Three focused prompts consistently beat one vague, large request.
Can AI check its own work?
Yes, with specific critique instructions. The key is giving the AI a structured critique role and explicit evaluation criteria - not just “make this better.” A good critique prompt quotes specific problems, explains why they’re issues, and suggests concrete fixes. The Plan-Draft-Critique method builds this self-review into the workflow systematically.
Why does AI give better answers with smaller prompts?
AI models have limited attention. When you give a model a massive prompt with multiple objectives, quality drops on each part because it’s spreading processing across too many goals simultaneously. A focused prompt with one clear objective gets the model’s full “concentration.” This is why the Plan-Draft-Critique method splits complex work into phases.
What is multi-step prompting strategy?
Multi-step prompting is the practice of breaking a task into sequential prompts rather than one large request. Each prompt handles one aspect of the task, and outputs flow from one step to the next. The Plan-Draft-Critique method is a specific multi-step prompting strategy that structures the steps as planning, writing, and reviewing phases.
Get PluggedIn
Stop rebuilding your Plan-Draft-Critique prompts from scratch every time you start a new project type.
Every complex project you tackle without a ready prompt chain costs you another 30 to 60 minutes of setup before you write a single useful word.
Get PluggedIn to go from rebuilding prompt chains by trial and error each session to dropping in a pre-built chain and getting usable output on your first attempt
What’s inside the Prompt Engineering Mastery Bundle:
Complete 9-lesson ebook (PDF)
7 niche-specific prompt packs (55+ prompts):
Customer support automation
Content creation on a budget
Client proposals & SOWs
Research & analysis
Email & communication
Sales & lead nurture
Operations & SOPs
What Comes Next
You now know how to break big tasks into manageable stages that AI can handle reliably. In the next lesson, we’ll explore how to ground AI models with your own sources using RAG (Retrieval-Augmented Generation). You’ll learn how to add context from your documents so answers stay close to facts, how to chunk content properly, and how to write prompts that cite sources correctly.
The difference between amateur and expert AI use? Experts don’t ask AI to do everything at once. They stage the work, guide each step, and stay in control from plan to final output.
Thank you so much for breaking everything down and explaining each part. This is where I’ve been stuck like three or four months. I couldn’t figure it out. But walking through it and doing it makes a whole lot of difference! Please do more.!