The Master's Review

Split the review into its own turn: paste your draft prompt and ask what is ambiguous or self-contradictory in it before asking for any output. Reasoning models are good at this because they are already checking their work internally - the thing they cannot do is check it against the world, so keep the review aimed at structure, logic and ambiguity rather than at facts. A standing line in Custom Instructions - "if my request is ambiguous, ask before answering" - gets you a lighter version of this review on every prompt without having to request it.

• Review the prompt in its own turn, before any output.
• Aim self-critique at structure and logic, not facts.
• Put "ask me if ambiguous" in Custom Instructions.
• Never trust it to catch its own factual errors.
• "Are you sure?" changes answers more than it checks them.

**Step 1 - review the prompt before running it.** Send this first, on its own:

I am about to send you a prompt. Do not carry it out. Review it.

---
PROMPT UNDER REVIEW: "Summarize the meeting notes. I want to know what the key takeaways were for the marketing and sales teams."
---

Tell me:

1. **Every way this could be read.** I think there are at least three: one combined list covering both teams, two separate lists, or one list of things that affect both. Which reading would you have picked, and why that one?
2. **What "key takeaways" means for each team.** Does marketing want strategy and insight while sales wants assigned tasks and deadlines? Those are different output shapes and my prompt asks for one word to cover both.
3. **What I have not told you**: how long, what format, whether to name the people who own each item, whether to include items where no decision was reached.
4. **The rewrite.** Give me the version you would rather receive.

Be blunt. If my prompt is fine, say so - do not invent problems to be helpful.

**Step 2 - the fixed prompt**, which is what the review above should get you to:

Analyse the meeting notes below. Produce exactly two sections, in this order.

## Key Takeaways for the Marketing Team
The strategic points, audience insights and positioning decisions relevant to marketing. Insight, not tasks.

## Action Items Assigned to the Sales Team
A numbered list of specific tasks. For each: the task, who owns it, and the deadline if one was given. Write "owner not named" or "no deadline given" rather than guessing.

Rules: only what the notes contain. If something was discussed without a decision, put it in a short third section called "Raised but unresolved" rather than writing it up as a takeaway.

NOTES:
[Paste the meeting notes here]

**Step 1 - get the prompt critiqued.** This is the cheaper order of operations: a weak prompt reviewed costs one turn, a weak prompt executed costs you a post you have to rewrite.

Act as a demanding editor reviewing a brief, not as a writer. Do not write the post.

---
BRIEF UNDER REVIEW: "Write a short blog post about why our new software is good."
---

1. **Name what is missing** that would change the output: who reads it, what the software does, which problem it solves, what a reader should do at the end, how long, what voice.
2. **Name the trap in the word "good".** What does a post built on that word look like, and why does nobody finish reading it?
3. **Ask me the five questions** you need answered to brief this properly - the five that most change what gets written, not a questionnaire.
4. Then rewrite the brief as you would want to receive it.

**Step 2 - the strong version**, once those questions are answered:

Act as a persuasive tech blogger. Write a 500-word post titled "3 Ways Our New Software Will Revolutionize Your Workflow". Energetic, benefits-focused.

For each of the three ways: name a specific pain point the reader has today, describe what it currently costs them in time or money, then show exactly how the software removes it. Pain first, product second - a reader who does not recognise their own problem in the first line will not reach the second.

**Then review your own draft.** After the post, add a short section headed "Where this is weak":

- Which of the three sections is thinnest, and what would fix it.
- Any claim I would need evidence for before publishing - any number, comparison or "faster than" that I have not given you. List these specifically. You cannot verify them and neither can your review of yourself; I have to.
- Anything you asserted about the software that I did not tell you. Flag it as invented rather than leaving it in the prose where it reads as fact.

That last point is the part worth having. A self-review will reliably tell you the second section is weaker than the first; it will not reliably notice that it made up a statistic in the third.

**Step 1 - ask for the contradiction to be named.** This is the clearest case for the technique, because the prompt contains an instruction that cancels another one and the writer cannot see it.

Do not carry out the prompt below. Audit it for internal contradictions.

---
PROMPT UNDER AUDIT: "Generate a list of 10 creative ideas for a new viral video. The ideas should be very unconventional and shocking. Make sure they are appropriate for a family-friendly brand."
---

1. **Point at the conflict.** Two instructions here cannot both be satisfied. Which two, and what happens when you try - do you get shocking ideas that break the brand rule, or safe ideas that break the creative one?
2. **Which one would you have obeyed?** Models usually resolve a contradiction silently by following one instruction and dropping the other, and the user never finds out which. Tell me which you would have dropped.
3. **Separate the real want from the wrong word.** I think I want "memorable and surprising" and I have written "shocking". Is that the actual conflict? Say so if you read it differently.
4. Rewrite it so both constraints hold.

**Step 2 - the resolved prompt:**

Generate 10 highly creative, unconventional ideas for a viral video. The ideas must be clever, surprising and heartwarming - memorable because they are unexpected, not because they are provocative. Everything must be completely appropriate for a family-friendly brand: no shock tactics, no controversy, nothing that depends on discomfort to be shareable.

For each idea: a one-line concept, why it would be shared, and which audience shares it.

**Then audit your own list.** Add a final section:

- Which three ideas are the most conventional, despite the brief. Be honest - "unconventional" is the instruction most quietly ignored, and five of these are probably variants of things that have already gone around.
- Which, if any, sit closer to the brand line than I would want. Flag them rather than removing them; that is my call.
- Any idea that resembles a campaign that already exists. Here you should say you cannot be sure - you are working from recollection, not a search, and this is exactly the kind of fact a self-review cannot establish. Name the ones worth checking and I will check them properly.