The Plan-Draft-Critique Method breaks a complex AI task into three separate prompts: Plan (outline the structure), Draft (write one section at a time), and Critique (review against specific criteria). Instead of one giant request, you stage the work in phases, review at each step, and catch errors before they compound into the final output.
What you will be able to do
- Write a Plan prompt that outlines structure only, using the instruction "Do NOT write the content yet, only the plan" to stop the model from jumping ahead
- Draft one section at a time from an approved plan, feeding in context from previous sections so quality and continuity hold up
- Write a Critique prompt that checks content against specific criteria (clarity, completeness, accuracy, consistency, audience fit) and prioritizes issues by impact
- Choose a list plan for straightforward sequential tasks or a tree plan for complex projects with dependent parts
- Use a debate prompt to argue two sides of a decision, with a recommendation, before committing to an approach
Before you start
- Access to an AI chat tool such as ChatGPT or Claude
- A task substantial enough to justify multiple prompts: the article's guide points to tasks over 500 words or with multiple distinct sections
- Willingness to review and approve output at each stage before moving to the next one
Reference
| Stage | Purpose | Key instruction or template element |
|---|---|---|
| Plan | Outline the structure before any content is written | "Do NOT write the content yet, only the plan." |
| Draft | Write one section at a time from the approved plan | "Write only this section. Do not continue to other sections." |
| Critique | Evaluate finished content against fixed criteria | Checks clarity, completeness, accuracy, consistency, audience fit; issues prioritized High/Medium/Low |
| List plan | Straightforward tasks with clear sequential steps | Numbered or bulleted list of items |
| Tree plan | Complex projects where parts connect to each other | Nested structure, e.g. 1.1, 1.1.1, showing dependencies |
| Debate | Decisions with no obvious right answer | Side A benefits/risks, Side B benefits/risks, comparison, recommendation |
| Refinement loop | Improve critical content over multiple passes | Plan, Draft, Critique, Revise, Re-critique, repeat until quality threshold is met |
Common errors and fixes
| What goes wrong | The fix |
|---|---|
| One prompt asks the AI to do the whole task at once (e.g. "Write a marketing plan for my new fitness app") | Split it into Plan, Draft, and Critique prompts instead of one giant request |
| Draft prompt has no scope limit | Model keeps writing past the current section, causing context drift; add "Write only this section. Do not continue to other sections." |
| Plan prompt is missing the no-content instruction | Model starts drafting content instead of just outlining structure; add "Do NOT write the content yet, only the plan." |
| Critique prompt just says "make it better" | Feedback comes back vague instead of actionable; give specific criteria and require quoted problems with concrete fixes |
| Single prompt used for a long or multi-section task | Output misses the target audience, ignores constraints, and wastes tokens on unusable content; use Plan-Draft-Critique when the task is over 500 words, has multiple sections, or is expensive to get wrong |
Read the full walkthrough
The complete lesson, with screenshots and any downloads, is published on Substack as part of Prompt Engineering for AI Automation.
More in this section
Continue the course
Browse all lessons in the Prompt Engineering for AI Automation course, or subscribe to the GenAI Unplugged newsletter to get new lessons in your inbox.
