The Master's Review

This is the tool for the half of the review that self-critique cannot do. A model asked to check its own facts has no way to look outside itself; an answer engine looks outside by construction and shows you the sources. So the honest division is: run the structural critique of your prompt in ChatGPT, Claude or Gemini, then bring the draft's factual claims here and ask which are supported, which are contradicted and which cannot be found. For reviewing a prompt's ambiguity it is the wrong tool and will simply answer the prompt.

• Use it for the fact check, not the prompt critique.
• Paste the claims as a list; ask for a verdict each.
• "Nothing found" is not confirmation - insist on that distinction.
• Open the citations; a link is not a vouch.
• Academic focus for research claims, Web for market ones.

**The honest note first.** Reviewing a prompt for ambiguity is not a research question, and asking it here will get you an answer about meeting summaries rather than a critique of your wording. Do that review in ChatGPT, Claude or Gemini, and do the summary there too - your notes are a private document and this tool is pointed at the open web.

What belongs here is the part a self-review provably cannot do. Meeting notes are full of claims about the outside world, and those are the ones that will embarrass you in a summary that gets forwarded.

Once you have the summary, bring its external claims here:

I have a set of meeting notes that make the following claims about things outside my company. For each, tell me what current sources say, cite them with dates, and give one of three verdicts: supported, contradicted, or not found.

1. [Claim about a competitor's launch, pricing or market position]
2. [Claim about a regulation, GST rule or compliance deadline]
3. [Claim about a market size or growth figure for our sector in India]
4. [Claim about an industry benchmark someone quoted in the meeting]

For each one:
- What do sources actually say, with the date of the source? A figure that was true two years ago is the most common way a meeting repeats itself into error.
- Is the original source the company itself, a regulator, a news outlet, or a consultancy report behind a paywall that everyone is quoting secondhand? That difference decides how much weight the claim carries.
- Where sources disagree, show the disagreement instead of averaging it.
- Where you find nothing, say "not found" explicitly. Do not treat the absence of a contradiction as support - that is the single most common misreading of a tool like this.

Link everything so I can open it. If a claim traces back to one press release that has been reprinted twelve times, tell me it is one source, not twelve.

**Wrong tool for the writing, right tool for two things around it.** Perplexity will not draft a persuasive blog post worth publishing - it answers with sources rather than writing with voice - and it will not usefully critique your brief either. Write and self-review the post in ChatGPT, Claude or Gemini.

But the review step there has a hole in it, and this is where the chapter's honest caveat bites: a model reviewing its own draft will tighten the argument and leave an invented statistic sitting in paragraph three. It cannot see the error because the error and the review come from the same understanding. Verification has to come from outside, and here is where it comes from.

**Use it first, for the pain points.** A post about pain points you guessed at is a post about nothing:

What are the documented, commonly reported workflow problems that [specific type of team - say, finance teams at mid-sized Indian firms] complain about with their current software? Cite sources from the last two years: user reviews on G2 or Capterra, industry surveys, forum threads, trade press. I want their words, quoted, not a summary of the category. Which complaints recur most often? Note where your evidence is thin.

**Then use it after, for the claims.** Paste the finished draft's assertions:

Here are the factual claims in a blog post I am about to publish. For each, a verdict - supported, contradicted, or not found - with sources and dates.

1. [Any time-saving or percentage claim]
2. [Any comparison with a named competitor]
3. [Any claim about what the market or the category does]
4. [Any statistic the draft quotes without a source]

For each: what the sources say, who published them and when, and whether the original is independent research or a vendor's own marketing. Where a number traces back to a single unsourced blog post, say so - that is how most software statistics propagate.

Flag separately any claim no source addresses at all. Those are the ones to delete rather than hedge, and they are precisely the ones the draft's own review reported as fine.

**Generate elsewhere; check originality here.** This tool will not brainstorm ten unconventional video concepts, and it will not spot the contradiction in your prompt - that review belongs in ChatGPT, Claude or Gemini, which will tell you that "shocking" and "family-friendly" cannot both be satisfied.

What it does better than any of them is answer the one question a self-audit cannot: has this already been done? A model reviewing its own list of viral ideas is working from recollection, and recollection about advertising campaigns is where confident wrong answers live. It will tell you idea 4 is fresh. It has no way to know.

**Research the ground first:**

1. Which Indian brand video campaigns from the last three years are documented as having gone genuinely viral - with reported view counts or press coverage, not just industry awards? Cite sources with dates.
2. For the family-friendly and heartwarming category specifically: which Indian campaigns in that register performed well, and what does trade press credit for it? I want the reported mechanism, not an inferred one.
3. Which Indian brand campaigns backfired for being provocative, and what was the consequence - complaints to the ASCI, withdrawal, boycott? Cite the cases with dates. This is the evidence for why the original prompt's "shocking" instruction was a mistake, and it is more persuasive than anyone's opinion about it.
4. What does the ASCI code actually say about content likely to cause offence? Link the current code rather than a summary of it.

**Then, once the ideas exist, come back with them:**

Here are 10 video concepts. For each, search and tell me whether a similar campaign has already run - in India or abroad, in the last five years. Verdict: closely resembles [named campaign, with link], loosely resembles, or nothing found.

Use Social focus as well as Web for this - a concept that ran as an influencer campaign may leave no trace in trade press. Be explicit about what you cannot establish: a small regional campaign may be entirely undocumented, so "nothing found" means not found, never "original". That distinction is the whole value of doing this step here rather than asking the generating model whether it was original.