Plan-and-Solve Prompting: Complete Guide with Examples (2026)

Last updated 2026-06-29 · Works with ChatGPT, Claude, Gemini
Quick Answer

Plan-and-Solve prompting instructs the AI to first devise a step-by-step plan, then execute it. The separation prevents the model from rushing to an answer before thinking through the approach. It consistently outperforms direct prompting on multi-step analysis, research tasks, and complex problem-solving.

What Is the Plan-and-Solve Framework?

Plan-and-Solve prompting was introduced as an improvement over standard Chain-of-Thought prompting. The key difference is the explicit two-phase structure: Phase 1 generates a plan (what steps to take), Phase 2 executes that plan (doing the actual work). This separation reduces errors caused by the AI skipping steps or losing track of the overall approach midway through execution.

Fields

T
Task
What needs to be accomplished. The more specific, the better the plan the AI generates.
P
Plan instructions
Tell the AI to create a step-by-step plan before doing any work. Example: First, devise a plan to tackle this. List each step before starting.
S
Solve instructions
Tell the AI how to execute the plan it created. Example: Now carry out each step in order, showing your work.
O
Output format
The structure of the final deliverable after execution is complete.
+R
Rules
Constraints: Show the plan before starting. Do not skip any planned step.

When to Use Plan-and-Solve

Use Plan-and-Solve for research tasks, competitive analysis, structured reports, debugging sessions, and any multi-step problem where jumping straight to an answer produces shallow results. It is especially effective when the task requires gathering or considering multiple pieces of information in a specific order. For simpler tasks, use RTF or RACE.

Examples

Competitive analysis
Task: Analyze the competitive landscape for a B2B prompt engineering tool targeting developers.
Plan instructions: First, devise a step-by-step plan for this competitive analysis. List every step you will take before doing any work.
Solve instructions: Now carry out each step of your plan in order. Show your reasoning at each stage.
Output format: Final deliverable: a comparison table (competitor, target user, key differentiator, pricing, weakness) plus a 3-sentence strategic recommendation.
Rules: Show the plan before starting. Cover at least 4 competitors. Use only publicly available information.
Bug investigation
Task: Investigate why our Express.js API returns 500 errors intermittently under load. Error logs show 'Cannot read property of undefined' in prompts route.
Plan instructions: Plan your debugging approach step by step before doing any analysis.
Solve instructions: Work through each step of your plan, analysing the likely causes at each stage.
Output format: List the top 3 most likely root causes ranked by probability, with a specific test to confirm or rule out each one.

Compared to Other Frameworks

Plan-and-Solve vs Chain-of-Thought: CoT asks the AI to show its reasoning; Plan-and-Solve explicitly separates planning from execution, which reduces step-skipping on complex tasks. Plan-and-Solve vs RISEN: RISEN gives the AI a fixed sequence of steps to follow; Plan-and-Solve asks the AI to generate its own plan first. Plan-and-Solve vs Skeleton-of-Thought: Skeleton-of-Thought outlines content structure; Plan-and-Solve plans a problem-solving process.

Save Plan-and-Solve prompts in Promptary — free

Save your first prompt in the structured editor and get a permanent REST API endpoint. Personal plan is free forever, no credit card required.

Start free

Frequently Asked Questions

What is Plan-and-Solve prompting?

Plan-and-Solve prompting instructs the AI to first create a step-by-step plan for a task, then execute that plan. The two-phase approach prevents the model from rushing to an answer before thinking through the approach, which improves accuracy and completeness on multi-step tasks.

How is Plan-and-Solve different from Chain-of-Thought?

Chain-of-Thought asks the AI to show its reasoning as it works toward an answer. Plan-and-Solve explicitly splits the process into two phases: generate a plan first, then execute it. The explicit planning phase reduces step-skipping and produces more thorough results on complex analysis tasks.

When should I use Plan-and-Solve?

Use Plan-and-Solve when the task has multiple distinct phases, when rushing to an answer without a plan produces incomplete results, or when you want the AI to cover a topic comprehensively before delivering the final output. It works especially well for research, analysis, and debugging tasks.

Does Plan-and-Solve work with Claude, GPT-4, and Gemini?

Yes. The technique is model-agnostic and works across all major language models. Models with stronger instruction-following tend to adhere to the plan-first constraint more reliably.