Thirty Use Cases, Five Operations. Learn The Five.
Thirty numbered entries, each one a verb and an arrow. RESEARCH / 30 sources → one answer. SUMMARIZE / 100 pages → what matters. REPURPOSE / 1 webinar → 20 assets. CLUSTER / 1,000 responses → themes.
Before anything else: this is the best-built card we have taken apart in this series. There are no product names on it, so nothing can be discontinued. No prices, so nothing can go stale. No links to dead domains. Every entry is a verb plus a transformation, which means it will read exactly as well in two years as it does today. And the footer is the most honest one we have quoted: “AI output needs human review. Always check facts, numbers and claims before you publish.”
So this post is not a correction. It is a compression. Because thirty is more than anyone will remember, and underneath the thirty there are only about five things actually happening.
But first, notice what every single arrow has in common. Look at the left-hand side of each one: 30 sources. Raw CSV. 100 pages. 50 calls. One webinar. 1,000 responses. Your messy spreadsheet. Your notes.
The input is always yours. Not one of these thirty asks Claude to know something about the world – no market sizes, no competitor pricing, no industry statistics. Every task is: here is my material, do something to it.
That is not a small detail. It is the entire reason this card holds up, and it is precisely what a business-strategy prompt pack we looked at last week got wrong – six of its eight prompts asked for facts the model does not have. Same brand, same format, opposite structure. This one is built the right way round.
The post this article is about
Read the original post: “30 ways to use Claude for marketing”
Thirty numbered use cases, footed “AI output needs human review. Always check facts, numbers and claims before you publish.” The tables below reproduce all thirty as printed.
Thirty entries, five operations
Sort the thirty by what physically happens to your material and they fall into five groups. The groups are not the same size, and they are not equally valuable.
Diagram: the five operations
| Operation | What happens to your material | Entries | What “human review” means here | Could you have done it yourself with more time? |
|---|---|---|---|---|
| 1. Compress many → few |
A large pile becomes a short answer | 7 — Research, Summarize, Extract, Interview, Meeting, Survey, Cluster | Did it drop something that mattered? Compression always loses something; the question is what | Yes – it saves hours, not judgement |
| 2. Expand one → many |
One thing becomes many variations | 8 — Repurpose, Personalize, Brainstorm, Script, Brief, Plan, Hook, Headline | Is this good, or just plausible? Volume is the easy part and the trap | Yes – and this is where most people spend all their time |
| 3. Clean messy → structured |
Disorder becomes a usable shape | 6 — Dedupe, Clean, Organize, Classify, Segment, Formula | Spot-check rows. A cleaning pass that silently mangles 2% is worse than no pass | Yes, slowly and badly |
| 4. Judge make it better by finding what is wrong |
Your work comes back with holes in it | 7 — Challenge, Critique, Validate, Simulate, Compare, Position, Edit | Is the criticism right? Push back – you know things it does not | No. You cannot be surprised by your own blind spot |
| 5. Compute numbers in, numbers out |
Figures get modelled | 2 — Calculate, Analyze | Check the arithmetic by hand. This is the row the footer is really about | Yes, and a spreadsheet does it more reliably |
Read the last column down. Four of the five groups save you time – real, substantial time, and that is worth having. Only one of them gives you something you could not eventually have produced alone.
You can summarise a hundred pages yourself if you have the afternoon. You can write twenty social posts from one webinar if you have the week. You cannot tell yourself what is wrong with your own strategy, because if you could see it, it would not be wrong.
All thirty, regrouped
Column 2 quotes the card exactly – the verb and the arrow as printed. We have only changed the order, putting like with like.
| Operation | The entries, exactly as printed | Our note – read 10 October 2026 |
|---|---|---|
| Compress 7 entries |
1. RESEARCH / 30 sources → one answer 3. SUMMARIZE / 100 pages → what matters 5. EXTRACT / 50 calls → recurring patterns 6. INTERVIEW / Calls → pains + objections 15. MEETING / Notes → actions + owners 17. SURVEY / Responses → key findings 29. CLUSTER / 1,000 responses → themes |
The most reliable group on the card. Entry 15 is the quiet winner – “actions + owners” turns a transcript into something with names attached, which is the difference between notes and a decision |
| Expand 8 entries |
7. BRAINSTORM / 5 directions, not 50 ideas 14. PERSONALIZE / One message → every audience 18. REPURPOSE / 1 webinar → 20 assets 19. SCRIPT / Idea → short-form video 20. HOOK / Generate + rank opening lines 21. HEADLINE / Generate + score alternatives 23. BRIEF / Idea → execution-ready brief 24. PLAN / Goal → tasks + deadlines |
Note how well entry 7 is written: “5 directions, not 50 ideas.” It is the only expand entry that asks for fewer outputs, and it is correct – fifty ideas is a way of avoiding a decision. Entries 20 and 21 both say rank or score, which quietly makes them judge tasks too |
| Clean 6 entries |
13. SEGMENT / Find audiences hidden in your data 22. DEDUPE / Messy list → clean data 25. ORGANIZE / Chaos → structured system 27. FORMULA / Describe it → Excel formula 28. CLEAN / Messy spreadsheet → usable data 30. CLASSIFY / Sort leads, feedback or content |
Entry 27 is the best-value item on the whole card – describing a calculation in English and getting the formula back is a task with a verifiable answer, because you can just run it. Entries 22 and 28 need spot-checking: a cleaning pass that quietly corrupts a few rows is hard to notice later |
| Judge 7 entries |
4. COMPARE / You vs 10 competitors 8. CHALLENGE / Make the case against your strategy 9. CRITIQUE / Ask what’s wrong, not what’s good 10. VALIDATE / Pressure-test before you build 11. SIMULATE / See it as your customer 12. POSITION / Find what only you can say 16. EDIT / Good → actually worth reading |
The highest-value block, and entries 8 to 11 run consecutively. See the section below. One caution: entry 4 needs you to paste the competitors’ actual material in – otherwise it is a fact question wearing a judge label |
| Compute 2 entries |
2. ANALYZE / Raw CSV → actual insights 26. CALCULATE / Model CAC, ROI + scenarios |
Entry 26 is the one to be careful with. Financial modelling written out in prose looks authoritative and is the hardest output on the card to eyeball. Have it show the formula, then compute it yourself in a sheet |
Entries 8 to 11 are four in a row that ask it to disagree with you
This is the part of the card worth framing and hanging up.
8. CHALLENGE – “Make the case against your strategy.”
9. CRITIQUE – “Ask what’s wrong, not what’s good.”
10. VALIDATE – “Pressure-test before you build.”
11. SIMULATE – “See it as your customer.”
Four consecutive entries, all doing the same unusual thing: instructing the model to push back rather than help. Entry 9 even names the failure mode directly – ask what’s wrong, not what’s good – which is a correction of how almost everybody actually uses these tools.
Because the default interaction is agreement. You paste your plan, you ask what it thinks, and you get a courteous, well-structured list of things that are promising about it. You feel validated. You learn nothing. The model was being helpful in the way it was asked to be.
These four flip the request, and that is why they are the only entries on the card that produce something you genuinely could not have made alone. A blind spot is by definition not visible from inside. Asking a fast, tireless reader to argue the opposite case is the cheapest adversarial review available – and unlike a colleague, it has no reason to be polite about your budget.
They are also the four nobody runs. Repurposing one webinar into twenty assets feels like a productive morning. Being told the strategy has a hole in it feels like a bad one. The feeling is backwards: the second one saves a quarter and the first one saves an afternoon.
We have arrived at this from the other direction more than once – an agent built from ten tutorials inherits their confidence, not their accuracy, and the missing step was always verification. Entries 8 to 11 are that step, written out, on a marketing card.
Learn the five and you can write the thirty-first
Thirty is a number you save and never open again. Five is a number you can hold while you work.
Once you know the five operations, you do not need the list, because you can generate any entry on it from the pattern. Every single one of these thirty is the same sentence with two blanks filled in.
| The pattern | Fill in these two blanks | A use case from the card | One the card does not have |
|---|---|---|---|
| Compress “Here is [a lot of X]. Give me [the few things that matter].” |
What pile · what shape you want out | “50 calls → recurring patterns” | “Here are 300 YouTube comments. Give me the five questions people keep asking.” |
| Expand “Here is [one thing]. Give me [N variations for N contexts].” |
The one thing · how many, and for whom | “1 webinar → 20 assets” | “Here is one video. Give me six thumbnail text options, each for a different viewer motive.” |
| Clean “Here is [something messy]. Put it into [this structure].” |
The mess · the exact structure you want | “Messy spreadsheet → usable data” | “Here is my channel’s video list. Tag each one by format, length band and topic.” |
| Judge “Here is [my work]. Tell me what is wrong with it, as [a sceptical someone].” |
The work · whose scepticism you want | “Make the case against your strategy” | “Here is my video plan for the quarter. Argue, as a subscriber who is bored of me, why they would not click any of it.” |
| Compute “Here are [my numbers]. Show the formula, then the result.” |
The numbers · and always ask for the formula first | “Model CAC, ROI + scenarios” | “Here is my revenue and view data. Show me the RPM formula, then compute it, so I can check it in a sheet.” |
The fourth column is the point. None of those four are on the card, and all four came out of the pattern in about a minute. A list of thirty is a sample of an infinite set, and the set is more useful than the sample.
Note what changes in the Judge row: you also have to supply whose scepticism. “Tell me what is wrong” gets you generic caution. “Argue as a subscriber who is bored of me” gets you something that stings, which is the point.
Four honest caveats
The footer on this card is the best one in the series. “AI output needs human review. Always check facts, numbers and claims before you publish.” Three categories, named specifically, in the right order of risk. Most cards manage a vague “check before you rely on it”; this one tells you what to check.
Our grouping is a reading, not a fact. Several entries sit on a boundary. HOOK and HEADLINE both say “rank” and “score”, which makes them half-judge. SEGMENT could be clean or compress depending on your data. COMPARE is a judge task that needs factual input pasted in first. We put each one somewhere so the table works; move them if your reading differs.
Two entries need more than the card implies. Entry 4 (COMPARE / You vs 10 competitors) only works if you paste the competitors’ actual material – otherwise it is asking the model for facts about ten companies, which is the failure mode we wrote about last week. And entry 26 (CALCULATE) produces prose that looks like a financial model; have it show the formula and compute it yourself.
And “entire marketing stack” is the caption’s word, not the card’s. The post promises you will stop switching between seven tabs. Thirty of these thirty operate on material you already have – they do not send an email, place an ad, or publish anything. This is a thinking stack, not an execution stack, and that is a compliment: a thinking stack is the half that was actually hard.
The one-line version
Thirty use cases, and every arrow starts with something you already own – your calls, your CSV, your webinar, your thousand survey responses. That is the structural reason this card is good and will stay good: nothing on it depends on the model knowing a fact about the world.
Underneath the thirty there are five operations: compress, expand, clean, judge, compute. Learn those and the list becomes unnecessary, because you can write the thirty-first entry yourself from the pattern.
And run entries 8 to 11. They are four in a row that ask Claude to argue against you, they are the only outputs on the card you could not eventually have produced alone, and they are the ones everybody scrolls past because being told you are wrong does not feel like progress.
Learning a lot does not make anyone good. Doing a lot does. Saving a card of thirty use cases is learning. Pasting this quarter’s plan in and asking it to argue the opposite – today, before you build any of it – is doing.
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.
Run entry 8 on something you are already committed to, and tell us what it found. Comments are under the original post – the objection it raised, not the fact that you ran it.
More on building a channel for a global audience at mmoyoutube.com.
