Lesson 21 of 32 · Part One, the principles

The Master's Review

Two reviews make the difference between a gamble and a plan. The first is yours: before you press send, read your own prompt as if you were the machine receiving it, and look for the word that could be read two ways or the detail you forgot. The second you ask for: have the model inspect your request for gaps before it answers, and inspect its own draft afterwards - through the eyes of the person who will read it.

Reach for it whenBefore anything that goes out of your hands - a message to an upset customer, notes you will revise from, a chapter you are about to send to a reader - and whenever your first instinct is to type "is this good?"

How to use The Master's Review in ChatGPT

Two app features turn the review from an idea into a habit. Canvas puts the draft beside the chat and takes suggested edits as comments on the lines they refer to, which is exactly the shape a memoir chapter or a long answer needs - you accept one, ignore the next, and the text stays in one place rather than multiplying into six copies. And editing your own message is how you test a reviewed prompt: when the model's questions reveal a gap, do not answer the questions below, go back and put the missing detail into the prompt itself, then run it again. The reviewed prompt is the thing worth keeping.

  • Work long drafts in Canvas so edits land on the lines themselves.
  • When its questions reveal a gap, edit the prompt, not the thread.
  • Treat its uncertainty list as a hint, never as an audit.

From GenAI for Everyone: Prompt Engineering Handbook, First Edition 2025, by Raja Rajeshwari Seshadri and Shreeram Iyer.

Two app features turn the review from an idea into a habit. Canvas puts the draft beside the chat and takes suggested edits as comments on the lines they refer to, which is exactly the shape a memoir chapter or a long answer needs - you accept one, ignore the next, and the text stays in one place rather than multiplying into six copies. And editing your own message is how you test a reviewed prompt: when the model's questions reveal a gap, do not answer the questions below, go back and put the missing detail into the prompt itself, then run it again. The reviewed prompt is the thing worth keeping.

From the book Open your draft in Canvas and ask ChatGPT to suggest edits; the comments appear beside the text, so you can accept or ignore each one. You can also edit your message and regenerate to test a reviewed prompt.
Worth knowingSelf-review is reliable on shape and tone, and not reliable on facts. Ask it to flag what it is unsure of and you get a list of things it happens to feel hedgy about - not an audit. A wrong section number under the GST Act, confidently cited, is precisely the kind of error that will not appear on that list, because the model has no way to look it up from memory. Treat the uncertainty list as a starting point and verify tax, legal and medical specifics against the source yourself. There is a second, subtler trap: asked to critique its own draft, it will often produce critique-shaped text - three polite observations, one of them invented - because that is what was requested. Push back and ask which single change actually matters.
  • Work long drafts in Canvas so edits land on the lines themselves.
  • When its questions reveal a gap, edit the prompt, not the thread.
  • Treat its uncertainty list as a hint, never as an audit.
  • Ask which one change matters most, to cut performative critique.
  • Verify every tax or legal specific against the official source.
The Unread Order vs. The Reviewed Request

Instead of

Explain GST input tax credit for my exam.

ChatGPT version

I am a CA Intermediate student in Surat.

**Before you answer anything**, review my request below and tell me what is unclear or missing. Ask up to three questions and stop there - do not answer and ask in the same message.

**My request:** one page of revision notes on GST input tax credit, covering the conditions for claiming it, blocked credits under Section 17(5), and the time limits. Short points, not paragraphs. One worked example with round numbers.

**After I have answered your questions and you have written the notes**, add a section headed "Verify these yourself" listing every section number, percentage and time limit you have stated from memory. I will check those against my study material and the CBIC site - so list them all, including the ones you feel confident about, because your confidence is not evidence.
Open ChatGPT 833 characters

Then loop it

  1. Good questions. My exam is in May, I know the basics already, so give most space to blocked credits.
  2. Now quiz me with five tricky MCQs from these notes. Put the answers at the end.

Why it worksIt makes the AI inspect the order for gaps before acting, and review its own answer for weak points afterwards.

"Is It Good?" vs. A Clear Editor's Brief

Instead of

Is my chapter good? Please check it.

ChatGPT version

Act as a kind but honest memoir editor. Below is the first chapter of my memoir, about 1,200 words, about my first teaching job in a village school near Kottayam in 1978.

Open it in Canvas and leave your suggestions as comments on the lines they refer to, so I can take them one at a time.

Review it for three things only:
1. Places where a reader may lose track of time or of who is who.
2. Sentences that have grown too long to follow.
3. Moments where I tell the reader a feeling instead of showing it.

Do not rewrite the chapter and do not improve my sentences. Up to ten numbered suggestions, each quoting the line it refers to. I write plainly; keep my voice.

[paste chapter]
Open ChatGPT 685 characters

Then loop it

  1. Thank you. I disagree with suggestion 4 - that headmaster really was that strict. Skip it, and show me how you would fix suggestion 2 only.
  2. Read your own suggestions again. Which three matter most for a first-time reader?

Why it worksA named set of review criteria turns a vague "check it" into a focused critique she can act on line by line.

The First Draft vs. The Self-Critiqued Draft

Instead of

Write a sorry message to an angry client.

ChatGPT version

I am a wedding photographer in Jaipur. A couple is upset: their album is three weeks late because the printing lab's machine broke down.

Write a WhatsApp reply under 100 words. It must apologise without hiding behind the lab, give the new delivery date of 20 November, and offer 10 extra prints free.

Then, in a second block headed "Read as the bride", review your own draft as if you were her, reading it on her phone while still angry. Quote any line that sounds defensive, cold or like a form letter, and say what she would think when she read it. Then give one improved version - not three.

Do not tell me the draft is good. If the whole thing needs rewriting, rewrite it.
Open ChatGPT 679 characters

Then loop it

  1. Better. The second version still says "unfortunately" twice. Remove that and make it warmer.
  2. Now give me a Hindi version I can send to the bride's mother.

Why it worksAsking the AI to critique its draft through the reader's eyes catches tone problems before the message is sent.

Questions

What is The Master's Review in prompt engineering?

Two reviews make the difference between a gamble and a plan. The first is yours: before you press send, read your own prompt as if you were the machine receiving it, and look for the word that could be read two ways or the detail you forgot. The second you ask for: have the model inspect your request for gaps before it answers, and inspect its own draft afterwards - through the eyes of the person who will read it.

When should I use this technique?

Before anything that goes out of your hands - a message to an upset customer, notes you will revise from, a chapter you are about to send to a reader - and whenever your first instinct is to type "is this good?"

How do I apply it in ChatGPT?

Two app features turn the review from an idea into a habit. Canvas puts the draft beside the chat and takes suggested edits as comments on the lines they refer to, which is exactly the shape a memoir chapter or a long answer needs - you accept one, ignore the next, and the text stays in one place rather than multiplying into six copies. And editing your own message is how you test a reviewed prompt: when the model's questions reveal a gap, do not answer the questions below, go back and put the missing detail into the prompt itself, then run it again. The reviewed prompt is the thing worth keeping.

What are the limits of this technique in ChatGPT?

Self-review is reliable on shape and tone, and not reliable on facts. Ask it to flag what it is unsure of and you get a list of things it happens to feel hedgy about - not an audit. A wrong section number under the GST Act, confidently cited, is precisely the kind of error that will not appear on that list, because the model has no way to look it up from memory. Treat the uncertainty list as a starting point and verify tax, legal and medical specifics against the source yourself. There is a second, subtler trap: asked to critique its own draft, it will often produce critique-shaped text - three polite observations, one of them invented - because that is what was requested. Push back and ask which single change actually matters.

How do I improve the answer after the first try?

With this principle, the loop begins before you even press Send. Read your prompt once as if you were the AI: is anything ambiguous or missing? Then, when the first answer arrives, ask the AI to review it - "What is weak or uncertain in your answer?" or "Check this against my original request." Read its critique, accept what is useful, overrule what is not, and ask for a revised version. A single review round usually lifts the quality more than any other follow-up.

Loop it

With this principle, the loop begins before you even press Send. Read your prompt once as if you were the AI: is anything ambiguous or missing? Then, when the first answer arrives, ask the AI to review it - "What is weak or uncertain in your answer?" or "Check this against my original request." Read its critique, accept what is useful, overrule what is not, and ask for a revised version. A single review round usually lifts the quality more than any other follow-up.

This lesson on the other tools

The technique is the same; the way you write it is not. Claude, Gemini.Perplexity.