The Master's Review

The original prompts were written here, so the review step needs little translation - what the platform adds is a clean division of labour. Keep a Gem configured as a prompt reviewer, so the critique arrives in a consistent shape every time instead of being requested afresh. Then use grounding with Google Search for the half of the review that self-critique genuinely cannot do: after the model has checked its draft for logic and structure, have it search for the factual claims it made and report what it can and cannot confirm.

• Keep a prompt-reviewer Gem for a consistent critique shape.
• Say "do not execute this prompt" or it will.
• Logic review first, then grounded fact check, separately.
• Demand openable links, not a confident summary.
• Put the review beside the draft in one Doc.

### PART 1: REVIEW THE PROMPT ###

Do not execute the prompt below. Review it. Nothing is attached yet, by design.

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

Review it against these points:

1. **Ambiguity.** How many different outputs would satisfy this? I count three readings of "takeaways for marketing and sales": one combined list, two separate lists, or only the overlap. Which would you have chosen?
2. **The hidden difference.** Marketing wants insight; sales wants assigned tasks. One word, two output shapes.
3. **Unspecified.** Length, format, whether to name owners, what to do with items discussed but not decided.
4. **The rewrite.** The version you would rather receive.

Do not search for this part - it is a question about my wording, not about the world. Do not invent faults to be useful; if it is adequate, say so.

### PART 2: THE FIXED PROMPT ###

I am attaching the meeting notes as a Google Doc. Work only from that file.

Produce two sections:

**Key Takeaways for the Marketing Team** - strategic points, audience insight, positioning decisions. Insight, not tasks.

**Action Items Assigned to the Sales Team** - numbered. Task, owner, deadline. Write "owner not named" or "no deadline given" rather than inferring.

**Raised but Unresolved** - anything discussed without a decision, so that open questions are not written up as conclusions.

Only what the document contains. Do not search - nothing from outside the file belongs in a summary of the file, and grounding is the specific way this goes wrong here.

### PART 3: THE REVIEW THAT NEEDS SEARCH ###

After the summary, list every item where the notes refer to something external - a regulation, a competitor's launch, a published figure, a deadline set by someone outside the company. For those only, now use Google Search and tell me with links whether each is current and correctly stated. Cite them so I can open them.

This is the division worth learning: you can review your own summary for structure, and you cannot review it for truth. The second part needs sources.

### OUTPUT ###

Into a Doc called "Meeting summary and review", with the review as a separate section at the end rather than mixed into the summary. Save Part 1 as a Gem called "Prompt reviewer" - I want this critique every time without retyping it.

### PART 1: CRITIQUE THE BRIEF ###

Act as a demanding editor. Do not write the post.

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

1. Why will a post built on the word "good" fail to hold a reader?
2. What is missing that changes the output: audience, what the software does, whose problem it solves, the action at the end, length, voice.
3. The five questions you need answered to brief this properly. Five that matter, not a form.
4. The rewritten brief.

No search for this part.

### PART 2: WRITE IT ###

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

For each of the three ways: the pain point the reader has today, what it costs them in time or rupees, then exactly how the software removes it. Pain first, product second.

Write it into a Google Doc called "Software post draft" so I can edit in place.

### PART 3: TWO SEPARATE REVIEWS ###

Keep these apart - they have different reliability and mixing them hides which is which.

**Review A - what you can judge yourself.** Which of the three sections is thinnest and what would fix it. Where the pain point is generic rather than specific. Whether the ending actually asks the reader to do something. This review is worth having.

**Review B - what you cannot judge yourself, and must search for.** List every factual claim in the draft: any number, any comparison, any "faster than", any claim about what competing tools do. For each, use Google Search and report one of three verdicts with a link I can open: confirmed, contradicted, or nothing found. Where you find nothing, say nothing found - do not promote an absence of contradiction into support.

Also flag separately: anything you asserted about our software that I never told you. That is invention, not error, and search will not catch it because there is nothing to find.

The point of splitting these: a self-review will reliably find that your second section is weaker than your first, and will reliably walk past a statistic you made up. Only Review B catches the second kind, and only if the links are real.

### PART 1: AUDIT THE PROMPT ###

Do not generate ideas. Audit this for internal contradiction.

**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. Two instructions here cannot both be satisfied. Name them.
2. Which would you have followed and which would you have silently dropped? That is the answer I want - a contradiction gets resolved invisibly and I never find out which half went.
3. Is this a real conflict or a wrong word? I may mean "memorable and surprising" and have written "shocking".
4. Rewrite it so both constraints hold.

No search. This is about my wording.

### PART 2: THE RESOLVED PROMPT ###

Generate 10 highly creative, unconventional ideas for a viral video. Clever, surprising, heartwarming - memorable because unexpected, not because provocative. Completely appropriate for a family-friendly brand: no shock tactics, no controversy, nothing needing discomfort to be shareable.

Per idea: one-line concept, why it gets shared, who shares it.

### PART 3: THE SELF-AUDIT, AND THE PART THAT NEEDS SEARCH ###

**What you can audit yourself:** the three most conventional ideas on your own list, despite the brief. "Unconventional" is the constraint most quietly ignored. And anything sitting closer to the brand line than I would want - flag it, do not remove it.

**What you cannot:** whether these ideas already exist. Now use Google Search on each of your ten and tell me, with links, whether a similar campaign has run - in India or elsewhere, in the last three years. Cite them so I can watch the videos.

This is the honest limit of the technique and this example shows it plainly. Reviewing your own list against my brief is something you can do. Knowing that idea 4 is a Zomato campaign from two years ago is not - without a search you are working from recollection, and recollection about advertising is exactly where a confident wrong answer comes from. Where search finds nothing, say nothing found rather than concluding the idea is original.

### OUTPUT ###

A Google Sheet called "Viral video ideas" - columns: Idea, Why shared, Who shares, Conventional (yes/no), Existing campaign found, Link. Leave a blank column for my own score.