Five Scores Out Of Ten. It Can Only Calculate Three.

One prompt this time, not a deck. “Act as a Creator Monetization Strategist, Business Model Expert, and Digital Product Consultant. Help me discover multiple ways to monetize my content niche without relying on just one income stream.”

Then eight labelled brackets to fill in — niche, audience, skills, platforms, audience size, existing products, budget, income goal — and a footer that says, in grey: “Replace every [bracket] before you send it.” Ten opportunities out, seven fields described for each, five scores from one to ten, three finalists and a testing plan.

It is a well-made prompt. The interesting part is what happens to those five scores.

Three of the five scores are about you. Two are about a market.

The rubric is: “Score each opportunity from 1–10 for demand, profitability, ease of execution, scalability and fit.”

Those five are not the same kind of thing at all, and the difference decides which of them means anything.

Ease of execution is a question about you. Can you build this, with the skills and the budget you listed four lines above? The model has those in front of it. It can reason about them.

Fit is the same. Does this match your niche, your audience, the platforms you named? Again: supplied, in writing, by you.

Scalability is a property of the business model itself, and it is close to definitional. One-to-one consulting does not scale; a recorded course does; a membership sits between them. That is structural knowledge, the kind a well-read model genuinely has.

Demand and profitability are different. Those are measurements of a market — how many people want this, at what price, with what margin. They live in search tools, in competitor pricing pages, in your own sales data. A language model has none of them, and nothing in those eight brackets supplies them.

So when the output arrives, you get something like Demand 8 · Profitability 7 · Ease 6 · Scalability 9 · Fit 8. Five numbers in a row, identically formatted, three of them reasoned from your inputs and two of them produced because a number was requested. Nothing in the formatting tells you which is which.

Diagram: what each score is a property of

Why mixing them is worse than leaving them out

If the two unknowable scores simply came back as a shrug, you would ignore them and use the other three. That is not what happens. They come back as 8 and 7, and then you do what anybody does with five numbers on the same scale: you compare rows, and the comparison is driven by whichever numbers vary most.

Two things follow, and both are quiet.

The guessed scores contaminate the reasoned ones. An idea that genuinely fits you and that you could genuinely build — a nine on fit, an eight on ease — loses to one the model happened to rate high on demand. You have let an unmeasured number outvote two measured ones.

And the ranking is the only part most people read. The prompt ends by asking for the three strongest options. Those three are selected by the composite, so the two scores with nothing behind them get a vote on which ideas survive to the action plan.

The fix costs one sentence, and it preserves everything good about the rubric:

“Score only ease of execution, fit and scalability from 1–10, and justify each score by quoting the information I gave you. For demand and profitability, do not give a number. Instead state what you would need to know, and give me the exact search terms, competitor pages or price points I should check myself.”

Same prompt, same ten ideas, but now the output separates what was reasoned from what must be verified — which is the discipline we found in a research deck that gave the model a legal way to say “I don’t know”. Here you are doing the same thing to a scoring column.

The seven output fields, and the one that decides everything

The rest of the output spec is better than the scoring, and one field in particular is the reason to keep this card at all.

That last field is the one almost every monetisation list leaves out. Generating ten revenue models is easy; naming where customer number one comes from is where nine of them quietly die. “Sell a course” is an idea. “Sell a course to the forty people who replied to your pinned comment asking how you do your thumbnails” is a plan. Asking for it by name in the output spec is the single best design decision on this card.

And it pairs with the sharpest line in the left column, which is easy to skim past: “identify the problems, desires and needs my audience is willing to pay to solve.” Not what interests them. Not what they engage with. What they pay for. Those are three different lists, and most creators conflate the first with the third for years. We have written the long version of that as a channel does not pay you — the business behind it does.

It asks for your audience size and then never uses it

Current Audience Size: [INSERT SIZE] is the fifth bracket. It is the only number in the whole input block that constrains what is physically possible, and nothing downstream touches it — least of all the final instruction, which is to take the three strongest options and “create a simple action plan for testing each one.”

Testing three offers at once is a reasonable thing to do with a large list. With a small one it is the fastest way to learn nothing, because the arithmetic does not work. Here it is, with our assumptions on the table so you can change them.

So the honest version of the final instruction, for most people reading this, is “identify the three strongest and test them one at a time, in this order, with a stated stopping rule for each.” Sequence beats parallel at small scale, every time, because the only thing you actually get back is attributable.

This is also why the audience-size bracket deserves to be load-bearing rather than decorative. At five hundred people the right answer is usually not a new product at all — it is one conversation with the ten most engaged people. At fifty thousand it is a different question entirely. The relationship between who your audience is and what they will pay matters more than which of ten models you picked.

What this card gets right

Three things, and they are the reason to keep it rather than bin it.

• Eight named brackets. Not one vague “[INSERT DETAILS]”. Niche, audience, skills, platforms, size, existing products, budget, goal — each with its own label, each a thing only you know. Compare that with the card we looked at yesterday, where the entire personal input was a single bracket and the model was asked to rank a market instead.
• “Avoid generic monetization advice.” A closing instruction that explicitly rejects the default output. Whether it works is another matter, but it names the failure mode.
• “Replace every [bracket] before you send it.” In the smallest type on the card, and it is the whole mechanism. A template with unfilled brackets does not produce a worse answer — it produces an answer about a generic creator who does not exist.

Four honest caveats

1. We did not run the prompt. This is a reading of what each requested field needs against what the input block supplies. The three construction fields may well produce genuinely good ideas; nothing here says otherwise.
2. The response-rate table is illustrative arithmetic. Ten per cent reach and two per cent response are round numbers chosen to show the shape, not measured benchmarks for your channel. Put your own in. The conclusion — that splitting a small sample three ways makes the result unattributable — holds at any plausible pair of numbers.
3. “Demand” is not always unknowable. If you paste in your own sales history, a competitor’s public pricing, or survey replies from your audience, the model can reason about demand perfectly well — because you supplied the evidence. The objection is specifically to scoring it from an empty input block.
4. Nothing here is business advice. It is an argument about which numbers in a prompt’s output were computed and which were produced on request.

The one-line version

A good template with one soft spot: it asks for five scores on the same scale when it can only work out three of them, and the other two get a vote on which ideas reach your action plan. Split the rubric — score ease, fit and scalability; for demand and profitability, ask for the search terms instead of the number. Then take the three winners and test them one at a time, because the audience-size box you filled in at the top is the number that decides whether testing three at once can tell you anything at all.