Hey, Mark here.
Last Wednesday a client who runs listing coaching for real estate agents asked me for hooks. A lot of them. He wanted to fill a month of ad tests, so I opened Claude and typed the thing everyone types. "Give me 20 ad hooks for this offer."
I got twenty. I read them on the second cup of coffee, and by number seven I already knew how the rest would go.
Twenty hooks. One idea, twenty coats of paint. "Tired of cold calling?" and "Sick of cold calling?" and "What if you never had to cold call again?" all sitting in the list like they were different.
I've done this for years and I still reached for the lazy prompt.
So this is the system I built to stop reaching for it, and it turns out the fix is older than Claude.
Let me show you the build.
The Hook Matrix
Here's what's happening when you ask for twenty hooks and get one:
Claude has no map of the territory. You asked for twenty of something without telling it how many kinds of that something exist.
So it finds the most obvious angle for your offer, writes it, and then spends the next nineteen tries paraphrasing the thing it already said.
It's doing exactly what you asked, which was quantity with no instruction about range.
A hook has one job. Stop the scroll and earn the next line. There are only so many psychological doors you can knock on to do that, and they have names. Fear. Curiosity. A number. A confession. A common enemy.
The doors have been the same for a hundred years of direct response, and a working copywriter carries the list in their head without thinking about it.
Claude doesn't carry that list unless you hand it over.
When you give it the map first, "give me twenty hooks" becomes "give me hooks across twelve different psychological angles," and the sameness has nowhere to hide. You also get something better than variety, which I'll come back to at the end.
The build has three steps:
Load the map.
Read the spread it produces.
Drill the angles that fit.
Let’s take a look.
Why Claude Gives You the Same Hook Twenty Times
Ask a model for twenty variations and it optimizes for fluency, not spread. It generates the highest-probability hook for your offer, and then every subsequent hook is anchored to that first one, because that's how the generation works.
The result reads like range but it isn't. It's one angle, rephrased until the list is long enough.
The first hook for my agent-coaching client was "Stop wasting time on cold calls." Numbers two through twenty were "Cold calling is killing your business," "Ditch the cold calls for good," "Cold calls are costing you listings."
Every one of them is the same door.
Range is a different instruction than volume. And the model will only give you range if you define what the range is made of.
The Twelve Proven Hooks To Start With
Twelve angles, each one a different psychological door into the same prospect. I keep the list to twelve on purpose. It covers the ground without turning into a reference manual nobody opens.
The examples below are all written for one sample offer, that real estate listing-coaching program, so you can see twelve different angles on identical material.
They're illustrations, not claims to reuse.
# | Hook type | What it does | Example |
|---|---|---|---|
1 | Direct Callout | Name the exact prospect and their moment, so the right person feels spoken to | "Agents who haven't taken a new listing since the spring rush: this one's for you." |
2 | Contrarian | Reject the advice the market keeps hearing, then promise the better way | "Cold calling is the slowest way to win a listing in 2026." |
3 | Open Loop | Start a story you don't finish, so they read on to close it | "The listing that rebuilt my year came from a woman who'd hung up on me twice." |
4 | Statistic | Lead with one specific, surprising number | "Top-producing agents talk to eleven people to win a listing. Everyone else talks to two hundred." |
5 | Question | Ask the thing already keeping them up at night | "What happens to your pipeline the month the agent down the street starts running ads?" |
6 | Command | A blunt imperative that interrupts the scroll | "Stop buying portal leads. You're renting clients who were never yours." |
7 | Named Mistake | Point at a specific error they're probably making | "Most agents lose the listing in the first ninety seconds of the appointment, for the same reason every time." |
8 | Big Promise | State the outcome plainly, with a timeframe | "Three listing appointments in the next fourteen days, without a single cold call." |
9 | Curiosity Gap | Tease the mechanism without naming it | "There's a two-line text that turns a dead lead into a listing appointment." |
10 | Common Enemy | Name the thing you and the prospect are both against | "The portals want you dependent on their leads. Here's how the top agents cut them out." |
11 | Confession | A first-person admission that lowers the guard | "I spent six years sure that cold calls built my business. One referral I almost ignored proved me wrong." |
12 | Future Pace | Put them inside the after-state | "Picture the Monday your listing appointments are already booked and you didn't dial a soul to fill them." |
Twelve doors, not one door twenty times. Some of them are clearly wrong for this offer, which is exactly what you want, because the ones that feel wrong are information too.
Load the Taxonomy Into Claude Once
You teach Claude the map a single time, then reuse it on every offer. If you run Claude or ChatGPT Projects, this goes in as a project instruction or a knowledge file and you never paste it again. If you don't, it's prompt one in a fresh chat.
You're a direct response copywriter who writes ad hooks. A hook is the opening line or two of an ad, and its only job is to stop the scroll and earn the next line.
You work from a fixed taxonomy of twelve hook types. Every hook you write is one of these, and you will tag each hook with its type in brackets.
1. Direct Callout — name the exact prospect and their moment so the right person feels spoken to.
2. Contrarian — reject the advice the market keeps hearing, then point to a better way.
3. Open Loop — begin a story or situation you don't resolve, so the reader has to continue.
4. Statistic — lead with one specific, surprising number.
5. Question — ask the thing already on the prospect's mind.
6. Command — a blunt imperative that interrupts.
7. Named Mistake — point at a specific error the prospect is probably making.
8. Big Promise — state the outcome plainly, with a timeframe.
9. Curiosity Gap — tease the mechanism or secret without naming it.
10. Common Enemy — name a shared enemy you and the prospect are against.
11. Confession — a first-person admission that lowers the reader's guard.
12. Future Pace — put the reader inside the after-state, already enjoying the result.
Rules:
- Every hook is tagged with its type, like [Contrarian].
- Write the way this specific market talks, not ad-speak.
- Do not resolve the tension inside the hook. A hook opens. The rest of the ad pays off.
- Keep each hook to one or two lines.
When you've read this, reply "Loaded" and ask me for the offer. Don't write hooks yet.
The taxonomy is the whole point, so it goes in before Claude sees the offer.
Load the offer first and the model starts writing before it has the map, which is the failure we're fixing. The bracket tag looks like housekeeping but it's the most important line in the prompt, and I'll show you why in the last section.
And the "reply Loaded, don't write yet" close keeps it from dumping twelve hooks before you've told it what it's selling.
Get the Spread, Then Read It as a Diagnostic
Now you give it the offer and ask for breadth, not polish. One or two hooks per type, all twelve types, in one pass.
Here's the offer.
- Product: [what it is, plainly]
- Audience: [who it's for, one specific sentence]
- Core promise: [the outcome they're paying for]
- Mechanism: [the reason it works when other things haven't]
- Proof: [any real proof — numbers, names, results. If you have none, say so.]
Write two hooks for each of the twelve types. Tag every hook with its type. Aim for range over polish. I want to see all twelve doors, even the ones that fit this offer badly.
You now have twenty-four hooks, and for the first time the list is actually a spread. Read it as a diagnostic before you fall in love with any single line.
Which types produced something with a pulse? Which ones came back limp?
For the agent-coaching offer, Contrarian, Common Enemy, and Confession all had teeth, because the market has a real villain (the lead portals) and a real shared story (everyone hated cold calling).
Statistic came back weak, because the client had no hard number worth leading with. That's the offer telling you where its angles live.
Ask for that read directly.
Before I pick any, tell me which three hook types fit this offer best and which three fit worst, and why, based on the audience and the promise. Two sentences each.
That question has saved me more time than any single hook. It turns a pile of copy into a map of where this particular offer is strong, and it points you at the two or three angles worth going deep on.
Drill the Types That Fit, Twenty at a Time
Here's where the volume you actually wanted shows up.
Take a type that had a pulse and drill it. Twenty of that one type, and you'll get real variety inside the angle instead of twenty paraphrases across all of them.
Give me 20 more [Contrarian] hooks for [product name]. Keep the core promise in every one, and keep saying [product name] so you don't drift onto a different product. Push for genuinely different angles inside the Contrarian frame, not rewordings of the same sentence.
The two instructions in the middle matter more than they look. Repeating the product name stops the classic drift where Claude wanders off and starts writing hooks for some adjacent product it invented.
And "different angles inside the frame, not rewordings" is the same anti-sameness instruction from the top, applied one level down, because the paraphrase habit shows up inside a single type too.
Run it for each type that earned it. Two or three types, twenty each, and you've got sixty aimed hooks with a label on every one, generated in the time the blind prompt took to give you twenty useless ones.
The labels are also a testing variable. Ship every hook tagged with its type, log the result next to the tag, and after thirty or forty tests you stop guessing which angle your market answers to.
On the agent-coaching account, Contrarian hooks beat Big Promise on click-through in eleven of fourteen tests. I can't hand that to a supplement client as a rule. It's a prior I built for this one market, and it only exists because the type was written down next to the number.
Volume without the label teaches you nothing. Volume with the label compounds.
Track Which Type Wins and Feed It Back
A winning hook you can't explain is luck. You got a good line, you don't know why, and you can't do it again on the next account. A winning hook with a type label is a strategy.
When "Confession" keeps beating "Big Promise" for this audience, you've learned something about the market that outlives the single ad, and your next batch starts from that knowledge instead of from zero.
Generate across the twelve, test, and let the winners tell you which doors this market opens for. Then drill those doors harder next round. The taxonomy that gave you range on day one gives you a sharpening feedback loop by month three.
That's the difference between generating hooks and building an actual read on a market. Same prompts. One extra habit, which is writing the type down.
Where the Labels Stop Helping
Three honest limits before you run this on live spend.
The taxonomy names the angle, never the execution. A hook tagged [Curiosity Gap] can still be a bad curiosity gap. The map tells you which door you're knocking on. It doesn't guarantee anyone answers. Your judgment on whether a specific line actually stops the scroll is still the job, and the label can make a weak hook feel more finished than it is.
Statistic hooks need a fact-check every time. Ask Claude for a number-led hook when your offer has no number, and it will invent one, confidently, the way it invents anything you push it toward that isn't there. Nova ran a whole test on this in the last issue and watched the model manufacture a survey stat out of nothing. Every number in a Statistic hook comes from the client's real records or it doesn't ship.
Big-promise hooks need a compliance pass per platform. "Three appointments in fourteen days" is fine in an email and can get an ad account flagged, depending on the platform and the category. The taxonomy will happily generate promises your ad account can't run. That edit is yours, and it comes before the hook goes anywhere near a media buyer.
None of that is a reason to skip the map. It's the reason the map ends at the point where your read on the market takes over.
The Master’s Memo
The blind prompt was never going to give you range that works.
Run the spread on your next offer before you write a single line yourself.
More clicks, cash, and clients,
Mark Masters
