Level 1 · Prompt Engineering

Make the AI Show Its Work as a Plan First

Ask the AI for a plan before the answer when the task has enough moving parts that a wrong assumption would waste your time.

Start With The Plan, Not The Polished Answer

When you ask ChatGPT or Claude for a finished answer right away, it has to guess the route and the destination at the same time. That can work for small tasks, like rewriting one sentence. It breaks down when the task has parts, priorities, or hidden assumptions, such as planning a trip, drafting a sensitive email, comparing options, or researching a purchase.

A plan-first prompt means you ask for the structure before the final result. Try: "Before writing the email, give me a short plan: goal, audience, key points, tone, and anything you need to assume." This turns the AI's thinking into something you can inspect. You are not asking for mystical reasoning. You are asking it to show the checklist it intends to follow, so you can correct it before it writes confidently in the wrong direction.

Use Plans When Mistakes Would Be Expensive

The more a task depends on context, the more useful a plan becomes. If you ask, "Help me prepare for a meeting with my manager," the AI may assume you want to negotiate, report progress, or solve a conflict. A quick plan exposes that assumption. If the outline says "ask for a raise" and the meeting is actually about project priorities, you can fix the path before the final draft exists.

Use a plan first for emails that affect relationships, research where sources may conflict, decisions with tradeoffs, and writing that must match a purpose. Skip it for simple transformations: "make this shorter," "turn this into bullets," or "fix the grammar." The tradeoff is speed. A plan adds one extra step, but it prevents the slower problem: editing a polished answer built on the wrong idea.

Ask For The Right Kind Of Plan

Once you decide to plan first, choose the shape that fits the task. For writing, ask for an outline. For errands, events, or travel, ask for a checklist. For decisions, ask for criteria, options, risks, and a recommendation method. For research, ask what questions should be answered before giving advice. Plain labels matter because they force the AI to organize the work in a way you can judge.

A useful prompt is: "Make a brief plan first. Include the assumptions you are making, the steps you would take, and the final output you would produce. Do not write the final answer yet." For example, before a customer-service complaint email, ask it to plan the goal, facts to include, tone, and desired outcome. If the plan is too broad, say, "Cut this to five steps." If it misses something, add the missing constraint before asking for the final version.

Review The Plan Like A Smart First Draft

The plan is not automatically correct. Treat it as a first draft of the approach. Look for three common failures: it assumes facts you did not provide, it optimizes for the wrong goal, or it makes the task too big. If you ask for a weekend itinerary and the plan includes museums, but you hate museums, that is not a writing problem. It is a planning problem, caught early.

After you correct the plan, ask for the final answer using the approved version: "Use this revised plan and write the email." For higher-stakes work, add a quick review step: "Before finalizing, point out one weakness in this plan." This creates a simple sequence: plan, correct, produce, review. The central habit is small but powerful: make the AI reveal its intended path before you let it run.

Key takeaways

  • Ask for a plan first when the task has multiple parts, hidden assumptions, or real consequences.
  • Use an outline for writing, a checklist for action, criteria for decisions, and questions for research.
  • Tell the AI not to write the final answer until you approve or revise the plan.
  • Check the plan for wrong assumptions, missing constraints, and goals that do not match yours.
  • Skip the planning step for tiny edits where a wrong path is easy to fix.