Page 31 · Module 01, Think Like a Strategist · this screen continues that page

All 108 frameworks
1.4BeginnerModule 01 · Think Like a Strategist

The Refinement Loop

You get an AI output, it's okay but not quite right, and you don't know how to improve it

The prompt

Copy & paste ready
1.4The Refinement Loop
Step 1 — Draft: [Paste your original prompt and generate first output] Step 2 — Critique: Review the response you just gave. What are the 3 weakest parts? Where is it too vague or generic? Step 3 — Improve: Rewrite the original response with those 3 weaknesses fixed. Keep what was strong. Improve what was weak. Step 4 — Lock: Anything that would make this significantly better? If yes, apply it. If no, confirm this is the final version. Achieving AI Output Excellence Critique Ask AI to evaluate its work Draft Improve Generate initial AI Ask AI to fix output weaknesses
LabelsReplace these

LabelsReplace these

Pro tip, from the book

Use Step 2 as a standalone habit. After any important output- email, proposal, plan and ask AI 'What is the weakest part of this?' before you send it. Takes 10 seconds.

Who should run this prompt

Module role profile

Who should reason through this problem? The quality of your thinking is shaped by whose lens you borrow. A strategist and a risk analyst look at the same problem and see completely different things — both useful, both necessary. Before you prompt, choose who you need in the room. These profiles give you six internal perspectives and four outside views that challenge your assumptions before you commit to them.

The roles below are who the book suggests running this prompt — this lesson’s prompt doesn’t open with a single swappable role clause, so pick one and adapt the prompt’s own wording yourself.

Internal rolesThink from inside the organisation
Chief Strategy OfficerLong-term position, trade-offs, competitive moatSetting 3-year direction or evaluating a major pivot
McKinsey Senior PartnerMECE structure, evidence-first, no fluffStructuring a complex problem or board presentation
Product Strategy DirectorMarket timing, feature bets, customer signalDeciding what to build next and why
Strategy ManagerTranslating direction into quarterly prioritiesBreaking a big goal into executable 90-day plans
Research AnalystData quality, pattern recognition, no assumptionsValidating a hypothesis before acting on it
First Principles CoachStripping assumptions, rebuilding from fundamentalsWhen every solution feels like a variation of the same thing
Outside perspectivesChallenge your blind spots
Sceptical InvestorFinds the hole in every planBefore pitching or committing resources
Military Decision AnalystSpeed under uncertainty, clear action biasIn fast-moving situations with incomplete data
Devil's AdvocateArgues the opposite as hard as possibleWhen your team agrees too quickly
Socratic FacilitatorNever answers — asks until you find it yourselfWhen you need to think, not just receive an answer

◆◆◆◆◆ → ◆ seniority, board level down to specialist outside the organisation

How to make any role sharper

  1. Add years of experience: "...with 15 years in enterprise SaaS" produces different depth than just the title.
  2. Add what they care most about: "You care most about [X]" shapes every word of the output.
  3. Add their communication style: "Be direct. Flag risks first. No jargon." changes the tone entirely.
See also: 1.3 Role Prompting - The Expert Chair

Use this when

The problem it solves

without starting over. You move on, leaving quality on the table.

How the framework works

From the printed page

One prompt almost never produces the best result. The Refinement Loop is a four-stage iteration: Draft (generate first version), Critique (ask AI what's weak), Improve (ask AI to fix those weaknesses), Lock (confirm the final version meets your goal). Each loop takes under 60 seconds. Most people stop at Draft. They get an output that is 70% right and move on leaving the best version unbuilt. The Critique step is the most underused: AI is surprisingly honest when asked to evaluate its own work, surfacing gaps you might not notice yourself. Two loops is almost always enough

The method, in four moves

Do these in order
1

Generate your first draft with any reasonable prompt.

2

On the next prompt, ask AI to critique its own output specifically.

3

Ask AI to rewrite with those critiques applied.

4

Repeat once more if needed. Two loops is usually enough. Stop accepting the first output - iterate to excellence Most people run the critique step, get three weaknesses, fix them then stop. They skip Step 4 (Lock). That final check is where the last 10% of quality lives, and it's also where the error you didn't notice gets caught. Two loops takes four minutes. One loop leaves quality on the table.

Where the framework comes from

Iterative prompt design — design thinking principles applied to LLM workflows, 2023

When this page changes, your copy of the book does not have to.

The book prints the framework. This page keeps it — and keeps the prompt current, because prompting technique moves faster than print does. When it moves, this page is rewritten and the code in your copy keeps pointing at it. No second edition, no replacement purchase, no account.

If something here looks wrong — a prompt that does not match its framework, a figure that reads oddly — tell Raphy. That correction reaches every printed copy at once.

Framework1.4 of 108
Printed onPages 3031
Prompt extracted18 Sept 2026
Costs to readNothing, no account

Copying is free. An account keeps them.

Nothing on this page is behind a login and nothing ever will be. Save any prompt to your own list, pick up on another device, and send a colleague a link with your name on it. No card, no trial, no paywall on the prompts themselves.

See how it works
Save this framework to your own list
Pick up where you left off on another device
Send a colleague a link with your name on it