Six Of These Eight Prompts Ask For Facts The Model Does Not Have
A prompt pack this time, and a good one. Eight business-strategy prompts – growth plan, competitor teardown, market research, SWOT, full business analysis, pricing audit, go-to-market, customer segmentation. The cover slide promises Claude can now analyse your business “like a top MBA consultant”, and the eight prompts themselves sit in the comments.
They are genuinely well written. Each one names a role, states a task, lists the dimensions to cover, and ends by demanding a concrete deliverable – “the first 5 actions I should take, ranked by impact and effort”, “the 3 priorities that should receive the most attention”. That is far better than the one-line prompts that usually circulate.
So the useful thing to do is not to grade the writing. It is to sort them by what kind of thing they are asking for, because two very different kinds are mixed in here and they do not behave the same way at all.
Some ask the model to reason over information you supply: rank these actions, structure this into a SWOT, identify the binding constraint. That is what a language model is genuinely good at, and the answer is checkable because you supplied the inputs.
Others ask it for facts about the world: “Analyze the market size.” “Compare their pricing.” “Evaluate… willingness to pay.” Those are measurements. If no source is attached, the model will still answer – fluently, specifically, with numbers – and you will have no way of telling where the numbers came from.
Six of the eight ask for at least one thing in that second category.
The post this article is about
Read the original post: “8 powerful prompts to think like an MBA consultant”
The cover slide says “Swipe for the prompts”; the eight prompts are posted in the comments underneath. The tables below quote all eight as written.
Two kinds of question, one prompt style
Both kinds open the same way – “Act as a senior…” – and both produce an answer of the same length, in the same confident register. That is precisely the problem: the output gives you no signal about which kind of question you asked.
Diagram: reasoning versus facts
| REASONING questions | FACT questions | |
|---|---|---|
| What you are asking for | Structure, ranking, trade-offs, “what is the binding constraint” | Numbers and states of the world: market size, competitor pricing, acquisition channels, willingness to pay |
| Where the input comes from | You supply it in the brackets | It has to come from somewhere else – a search, a report, your own data |
| Is the model good at it? | Yes. This is the core competence | Only with a source attached. Without one it is reconstructing from memory |
| What a wrong answer looks like | A ranking you disagree with – visible, arguable, easy to push back on | A specific number, stated plainly, with no indication it was estimated |
| How you catch it | You already know the inputs, so you can see the reasoning | You cannot, from the answer alone. You have to go and check it |
| The fix | None needed | Attach a source, or require the answer to mark what is uncertain |
Row four is the one that costs money. If you ask for a growth plan and the ranking is wrong, you will notice, because you know your own business. If you ask for a market size and the number is wrong, it looks exactly like a number that is right – and a business plan built on an invented total addressable market is a plan that reads beautifully and points the wrong way.
This is the same shape as something we keep finding: an agent built from ten tutorials inherits their confidence rather than their accuracy. Fluency is not evidence, and a prompt that produces a polished report has not thereby produced a true one.
All eight, sorted by what they are asking for
Column 2 is the author’s own text, shortened only where it lists dimensions. Column 3 counts the placeholders – sixteen across the eight – and column 4 is the sort that matters.
| Prompt | Quoted from the prompt as written | Brackets | Asking for | Our note – checked 10 October 2026 |
|---|---|---|---|---|
| 1. Strategic Growth Plan | “Build a practical growth plan covering revenue levers, customer acquisition, retention, new markets, offers, pricing, and distribution. Identify the biggest constraint limiting growth… give me the first 5 actions… ranked by impact and effort.” | 4 [amount] [offer] [target] [timeframe] |
Reasoning | One of the best in the pack. “Biggest constraint” and “ranked by impact and effort” are reasoning tasks over inputs you supply. Four brackets means four chances to make it specific |
| 2. Competitor Teardown | “Compare their positioning, pricing, offers, strengths, weaknesses, customer acquisition channels, messaging, and key differentiators.” | 1 [list] |
Facts | Pricing changes constantly and acquisition channels are usually private. Paste the actual pricing pages in, or turn on search and ask for sources |
| 3. Market Research Report | “Analyze the market size, customer segments, demand trends… End with… the assumptions I should validate before investing.” | 2 [market] [offer] |
Facts (self-aware) | “Market size” is the single riskiest request in the pack. But this prompt is also the only one that ends by asking what to validate – see below |
| 4. Strategic SWOT Analysis | “Create a specific, evidence-based SWOT… Avoid generic observations. For every weakness, provide a practical corrective action; for every threat, a mitigation…” | 1 [business] |
Reasoning | The best-constructed prompt here. It is the only one that names its own failure mode (“avoid generic observations”) and forces every item to carry an action |
| 5. Full Business Analysis | “Analyze my business across business model, target market, competitive position, operations, revenue streams, and growth constraints… 3 highest-impact opportunities… priorities, risks, and key metrics.” | 1 [describe business] |
Reasoning (input-starved) | Six dimensions of analysis resting on one bracket. Write two lines there and you get a two-line-quality analysis dressed as a report |
| 6. Pricing Strategy Audit | “Evaluate my pricing based on customer value, competitors, target market, positioning, and willingness to pay.” | 2 [describe] [price] |
Facts | Willingness to pay is measured, not reasoned – from tests, interviews, or churn data. Without those the answer is a well-argued guess about your customers |
| 7. Go-To-Market Strategy | “Build a practical GTM strategy covering positioning, ICP, offer, messaging, acquisition channels, launch sequence… first 100 customers… milestones, metrics, and the biggest risks.” | 3 [product] [target audience] [market] |
Mixed | Planning and sequencing are reasoning. “Acquisition channels” for a specific market is a fact question hiding inside a plan |
| 8. Customer Segmentation | “Identify the 3 most valuable customer segments based on need, buying intent, revenue potential, and accessibility… pain points, buying triggers, objections, willingness to pay…” | 2 [describe] ×2 |
Facts | This one is only as good as your customer data. Paste real segments and it is excellent; describe them from memory and it will invent plausible ones |
The best sentence in the pack appears once
Prompt 3 ends like this: “Identify the most attractive opportunity, the biggest risks, and the assumptions I should validate before investing.”
That clause does something none of the other seven do. It asks the model to separate what it is confident about from what it is guessing, and it hands you a to-do list of things to go and check. It converts a report into a report plus a shopping list of unknowns.
It appears in exactly one of the eight prompts – and, awkwardly, in one of the prompts most likely to need it.
Prompt 4 does something adjacent with “evidence-based” and “avoid generic observations”, which is why it is the sturdiest of the set. The other six end by asking for conclusions and nothing else.
So here is the free upgrade, and it costs one sentence. Append it to all eight.
“Mark clearly which parts of this answer are based on information I gave you, which are general knowledge, and which are estimates. List what I should verify before acting, and tell me what you would need from me to make the weak parts stronger.”
That one addition changes the output from a document into a document with its own confidence labels. It is the same move we applied to a knowledge graph that collapsed three kinds of link into one number: when a tool knows the difference between extracted and inferred, make it say so.
Sixteen brackets, and they are the actual work
Count the placeholders across the pack and there are sixteen: [amount], [offer], [target], [timeframe], [list], [market], [business], [describe business], [price], [product], [target audience] and several more [describe]s.
Every one of those is a hole you fill, and the filling is where the entire quality of the answer is decided. Prompt 5 asks for an analysis across six dimensions – business model, target market, competitive position, operations, revenue streams, growth constraints – and gives you one bracket to describe the business in.
Write “I sell a course about YouTube” in that bracket and you will get a competent, generic report about selling courses. Write four hundred words – what you sell, to whom, at what price, how they find you, what your last three months looked like, where you are stuck – and the same prompt produces something worth reading.
The prompt is the cheap half. It is free, it is copyable, and everyone who saved the post now has the identical one – which is precisely why it cannot be the thing that makes the difference. We made the same argument about tool knowledge that everyone else also has. What you put in the brackets is yours alone.
How to run one of these properly
Pick the prompt that matches a decision you actually face this month. Then do these four things before you read the output.
| Step | What to do | Why | Time |
|---|---|---|---|
| 1. Overfill the brackets | Replace each placeholder with a paragraph, not a phrase. Numbers where you have them | This is the only input the model has about you. Sixteen brackets, and they decide everything | 20 minutes |
| 2. Paste the facts in, or turn on search | For prompts 2, 3, 6 and 8: attach the actual pricing pages, your actual revenue, your actual churn | Converts a fact question into a reasoning question – which is the kind it is good at | 15 minutes |
| 3. Append the confidence clause | Add the sentence above to every prompt in the pack | Makes the answer label its own guesses. The pack does this once; do it eight times | Seconds |
| 4. Ask one thing you already know | Include a question whose answer you are certain of, and check it | A confident wrong answer to that one tells you how to read the rest | 2 minutes |
| 5. Then read it as a draft | Treat the output as a structured argument to push back on, not a report to file | Prompt 4’s own instruction – “avoid generic observations” – applies to how you read it too | Ongoing |
Step four is the cheapest insurance available and almost nobody does it. Slip in a question about your own business that you can grade – last quarter’s best channel, your actual refund rate, a competitor’s real price – and see whether the answer is right. If it confidently gets that wrong, you now know exactly how much to trust the market-size paragraph.
Four honest caveats
These are well-written prompts and we would use several of them. Prompts 1, 4 and 7 are genuinely good: they name a role, list dimensions, and demand a ranked deliverable rather than an essay. Prompt 4 in particular is the sturdiest prompt we have quoted on this blog – it names its own failure mode and forces every finding to carry an action. The criticism here is about sorting, not quality.
“Act as a senior X” is doing less than it looks. All eight open that way. Role framing shapes vocabulary and structure, which is worth something – but it does not give the model information it does not have. An instruction to act like a market research analyst does not supply market research. It is the most copied line in any prompt pack and the least load-bearing.
The fact problem is solvable, not fatal. Every one of the six fact-heavy prompts becomes reliable the moment you attach a source: paste the competitor’s pricing page, upload the industry report, give it your own numbers. We are not saying do not ask about market size. We are saying do not ask about market size with nothing attached and then put the answer in a plan.
And a note on where we got these. The carousel says “swipe for the prompts”; the prompts themselves are in the comments under the post. We quoted all eight from there on 10 October 2026. If the author edits them later, our quotes will be the older version.
The one-line version
Eight prompts, two kinds of question. Reasoning questions – rank this, structure this, find the constraint – are what the model is for, and you can check the answer because you supplied the inputs. Fact questions – market size, competitor pricing, willingness to pay – will be answered with equal confidence and no source.
Six of the eight contain at least one fact question. All six become good prompts the moment you attach the facts yourself.
Then add the sentence the pack already uses once: mark what is an estimate, and list what I should verify. It is free, it takes the answer from a report to a report with its own confidence labels, and it is the difference between a strategy document and a strategy document you can act on.
And fill the brackets properly. Sixteen placeholders, and they are the only part of this that is yours – everyone who saved that post has the identical prompt.
Learning a lot does not make anyone good. Doing a lot does. Running all eight prompts in an afternoon produces eight documents and no decisions. Running one of them on a decision you actually face this month produces one decision, which is strictly more than eight documents.
If you want to be held to that, we run a free challenge at BANI Academy. There is no barrier to entry – you commit a small fee that is refunded when you finish, and the only requirement is that you actually do the work for the length of the challenge.
Which of the eight maps to a decision you are facing right now – and did you fill the brackets with a phrase or a paragraph? Tell us in the comments under the original post.
More on building a channel for a global audience at mmoyoutube.com.
4 Comments