Hey, Mark here.
Every copywriter I know keeps a folder of classic ads they admire.
We study them, we trade them in group chats, and then we close the folder and write from scratch anyway.
The structure we admired never makes it into the workflow. It stays a reference when it could be a tool.
This issue closes that gap. We'll build a reusable ad template from the 1962 Avis campaign, start to finish, and you'll walk away with both the template and the method for turning any classic you admire into the next one.
Let's take one apart.
Turn Any Classic Ad Into a Prompt That Writes New Ones
A classic ad you admire is a proven structure with the original content still attached. A template is what you get when you strip the content away and teach the structure to Claude.
The template is two prompts:
The first one teaches. It hands Claude a role, the framework extracted from the classic, and a few complete example ads that demonstrate the framework in different niches.
The second one executes. It collects the facts of your offer in a fill-in block and orders up a new ad that follows the structure the first prompt taught.
That's the whole machine: a teaching prompt and an execution prompt, built once per classic ad, reused on every project after.
Here's why the architecture produces usable copy when "write me a Facebook ad" doesn't. Instructions describe, but examples demonstrate, and Claude follows a demonstrated structure far more faithfully than a described one.
The example ads inside the teaching prompt are the real payload. The trade calls this few-shot prompting: you show the model two or three finished specimens and it learns the shape, the length, and the level of restraint from the specimens themselves.
The build has five steps:
Pick a classic ad that earned its run.
Extract its framework, then correct the extraction by hand.
Write the seed ads and assemble the teaching prompt.
Build the execution prompt with hard fields.
Add the gate that catches AI-sounding drafts.
We'll run all five on one ad. By the end you'll have a working template, and the second one you build will take half the time.
Step 1: Pick a Classic That Earned Its Run
Not every famous ad makes a good template. You're screening for three things.
It ran long enough to prove itself. Space in the classic era was expensive, and an ad that kept its slot for years was paying its way. Longevity is the closest thing the archives have to a conversion report.
Its structure is visible. You should be able to point at the ad and say "this line does this job." Ads that work through a single brilliant image or one untranslatable headline make poor templates, because there's no repeatable sequence to extract.
Its psychology transfers. The persuasive move has to work for offers that look nothing like the original product. Appeals tied to one era's social anxieties travel badly. Appeals built on how people weigh trust, proof, and risk travel fine.
Public archives like Swiped.co hold decades of candidates. And if you built the Swipe File Engine from a few issues back, you already have a personal library to query. Ask it for the pieces whose structure you admired enough to save, and pick from those.
For this build, the classic is Avis. The company had lost money for over a decade when this campaign launched in 1962. It turned a profit the following year, and the campaign ran for years after that. Here's the ad:
Avis is only No. 2 in rent a cars. So why go with us?
We try harder. (When you're not the biggest, you have to.)
We just can't afford dirty ash trays. Or half-empty gas tanks. Or worn wipers. Or unwashed cars. Or low tires. Or anything less than seat-adjusters that adjust. Heaters that heat. Defrosters that defrost.
Obviously, the thing we try hardest for is just to be nice. To start you out right with a new car, like a lively, super-torque Ford. And a pleasant smile. To know, say, where you get a good pastrami sandwich in Duluth.
Why? Because we can't afford to take you for granted.
Go with us next time. The line at our counter is shorter.
Read it twice: once as a reader, once as a mechanic looking for the moving parts.
Step 2: Extract the Framework, Then Correct It by Hand
Claude does the first pass at naming the moves. You do the second pass, and the second pass is what makes the template yours instead of generic.
Here's the extraction prompt. Paste the classic ad underneath it.
I'm going to paste a classic ad. I want to turn its structure into a reusable framework, so I need the structure, not the content.
Read it and give me three things:
1. The moves. List the ad's structural moves in order. For each move, name it in two or three words, quote the exact line or lines that perform it, and state in one sentence what the move does to the reader.
2. The engine. In one paragraph, explain why this structure persuades. What belief does it create, and what objection does it disarm?
3. The transfer conditions. What does a product or offer need to have for this structure to work on it? Name the conditions plainly. If an offer lacks them, this framework should be a bad fit.
Do not paraphrase the ad into modern copy. Do not add moves the ad doesn't make. If two moves happen inside the same sentence, list them separately.
[paste the classic ad]
Why it's built this way: Requiring a quote for every named move ties the analysis to actual lines, so Claude can't float off into generalities about "compelling storytelling."
The transfer conditions matter later: they go into the teaching prompt so the template can warn you when an offer doesn't fit, instead of forcing the structure onto everything.
And the last paragraph blocks Claude's favorite shortcut, which is skipping the analysis and rewriting the ad.
Now the correction: When I ran this on Avis, the first pass was plausible and slightly wrong. Claude labeled the middle of the ad a "benefit list" and praised its "confident, conversational tone." Both readings miss the actual mechanism.
Those middle lines are mundane on purpose. The ash trays, the gas tanks, the wipers: every detail is small, specific, and checkable within the first five minutes of a rental. A benefit list says "award-winning service." This ad says "seat-adjusters that adjust," and the smallness is what makes it believable, because a company faking its case would have reached for bigger claims.
That correction is a judgment call, and it's the kind of judgment you already have.
Here's the corrected framework, the version that goes into the template:
The admission. Open with a true, unflattering fact a competitor would bury. It has to cost something real.
The reader's question. Immediately ask the question the admission just raised: so why buy from us? You're steering the objection instead of waiting for it.
The reframe. Answer in one line that turns the flaw into the reason for a customer benefit. "We try harder. (When you're not the biggest, you have to.)"
The mundane proof stack. Three to five small, specific, checkable details that could only be true if the reframe is true. The smallness is the credibility.
The logic lock. One line explaining why this company has no choice but to deliver. "We can't afford to take you for granted." The reader finishes the thought themselves.
The underdog close. End on a benefit only the smaller player can offer. "The line at our counter is shorter."
The trade name for this shape is the damaging admission, and it's one of the most reliable structures in the archive. That's the extraction done. Now we teach it.
Step 3: Write the Seed Ads and Assemble the Teaching Prompt
The seeds are the example ads that live inside the teaching prompt, and they carry more weight than every instruction around them. Three rules govern them.
Write complete ads. Claude imitates whatever you show it. A half-finished seed teaches half-finished output.
Use unrelated niches. Two seeds from the same industry teach Claude the industry. Two seeds from different worlds teach it the shape, which is what you want, because next month's client won't look like this month's.
Write them at the standard you want back. The template inherits your seed quality permanently. This is where the afternoon goes, and it's the best-spent hour of the whole build.
One note on invention. My seeds below are fictional products, and that's fine, because a seed's job is to demonstrate structure. The facts in your real ads get protected in Step 4.
Here's the assembled teaching prompt, framework and seeds included. This is the first of the two files you'll save.
You're a direct response copywriter who writes short ads built on one specific framework: the damaging admission. You learned it from the 1962 Avis campaign, and you follow its structure exactly.
The framework has six moves, in this order:
1. The admission. Open with a true, unflattering fact about the product that a competitor would bury. It must cost something real. An admission that secretly flatters ("we're too popular") kills the ad.
2. The reader's question. Immediately ask the question the admission raises: so why buy from us?
3. The reframe. Answer in one line that turns the flaw into the reason for a customer benefit.
4. The mundane proof stack. Give three to five small, specific, checkable details that could only be true if the reframe is true. Use no superlatives. The smaller and more concrete the detail, the more the reader believes it.
5. The logic lock. Write one line explaining why this company has no choice but to deliver. The reader should be able to finish the thought themselves.
6. The underdog close. End with a call to action built on a benefit only the smaller or humbler option can offer.
This framework only fits offers that have a real disadvantage worth owning: smaller, fewer features, more expensive, slower, less famous. If the offer I give you has no true admission, say so and stop. Do not manufacture false modesty.
Here are two ads that execute the framework correctly. Match their length, their rhythm, and their restraint.
[Example 1 — project management software]
PlanBoard is missing about 200 features you'll find in the big project management tools.
So why would your team pay for it?
Because the tool your team actually opens beats the tool that can do everything. When you can't win on features, you have to win on Monday morning.
There's no Gantt chart nobody updates. There are no custom fields to argue about. There's no six-week rollout that quietly becomes someone's job. A new hire learns the whole tool before lunch. The status board is accurate on Friday afternoon because updating it takes less than a minute.
We can't afford to be complicated. Complicated is what you tolerate from software you're already locked into.
Start your next project in PlanBoard this afternoon. It may be the only software decision this year that doesn't need a meeting.
[Example 2 — wedding photography]
I photograph twelve weddings a year. Most photographers in this city book sixty.
So why book the one with the waitlist?
Because the photographer you meet at the consultation is the one who shows up on the day. When you only take twelve weddings, every single one has to be worth showing off.
There's no associate shooter you've never met. There's no gallery that arrives four months late, after the thank-you cards have already gone out. I walk your venue the week before. I learn the names of the six people who matter most before anyone lines up for photos. And your gallery is finished within two weeks, because yours are the only photos on my desk.
I can't coast on volume. At twelve weddings a year, every referral counts.
Three of my dates for next fall are still open. If one of them is yours, book the consultation this week.
[End of examples]
Both examples are fictional products, written to demonstrate the structure.
Read all of this, then reply with one word: "Ready." I'll give you the details of a real offer next.Why it's built this way: The role line is short because the framework and the seeds do the teaching, and a paragraph of "you are a world-class copywriter" adds nothing they don't.
The transfer conditions are in there so Claude flags a bad-fit offer instead of writing fake humility around it.
And the "reply Ready" close does two jobs: it confirms the whole prompt was read, and it keeps ad copy out of the first response, so nothing gets generated before your offer details arrive.
Step 4: Build the Execution Prompt With Hard Fields
The execution prompt is the fill-in block you'll complete for each new project. The design choice that matters here is the split between soft fields and hard fields.
Soft fields are Claude's to shape. The reframe wording, the voice, the rhythm of the close all improve with its judgment. Hard fields are facts, and Claude never supplements them. The admission, the proof details, and the offer terms come from you or they don't appear at all.
If you built the Swipe File Engine, you know this rule already. There it stopped Claude from inventing library entries. Here it stops your ad from inventing proof.
The second file to save:
Now write one new ad following the framework and the examples exactly.
Here's the offer. Fields marked HARD contain facts. Use them exactly as written. If a HARD field is blank, write the ad without that element. Never invent anything that would fill it.
- Audience: [who the ad speaks to, in one specific sentence]
- The admission (HARD): [the true, unflattering fact we're leading with]
- The reframe: [why that flaw produces a specific benefit for the customer]
- Proof details (HARD): [three to five small, true, checkable specifics that back the reframe]
- What we're selling: [the product or service, plainly]
- Call to action (HARD): [the one next step, plus any true scarcity or timing]
- Voice notes: [phrases this market uses, and phrases they'd never use]
Rules:
1. Follow the six moves in order. Add no greeting, no extra sections, no hashtags.
2. Write the way this audience talks to a trusted friend, at the reading ease of the two examples.
3. Keep paragraphs to one to three short sentences.
4. Every proof line must trace back to the proof details above. If you're tempted to strengthen one, don't.
5. Before you show me the ad, check it against three questions. Does each of the six moves appear once, in order? Does every specific claim come from a HARD field? Would a marketing director who knows this market believe a specialist wrote it? If any answer is no, revise once. Then show me the final ad only.
Why it's built this way: The blank-field instruction is the load-bearing line. Without it, an empty proof field becomes an invitation, and Claude fills it with a confident, fabricated specific that reads exactly like the true ones. With it, a thin brief produces a thinner ad, which is the honest outcome.
The voice notes field earns its place too, and the second half of it does most of the work: telling Claude what this market would never say kills more generic phrasing than any positive instruction I've found.
The bracket fields are there to be filled, every one of them, with real project detail before you run it. The quality of what comes back tracks the quality of what you put in this block more than anything else in the system.
Step 5: Make the Template Check Its Own Work
Rule 5 in the execution prompt is the gate, and it deserves a closer look because the details are deliberate.
The three questions check the three ways these drafts actually fail:
Broken structure, invented specifics, and AI-flavored prose.
The third question works because it names a reader with standards. Asked to "check your work," Claude grades itself generously. Asked whether a marketing director who knows the market would believe a specialist wrote it, it gets noticeably stricter about the lines that would embarrass you.
"Revise once, then show me the final ad only" caps the loop. Left unbounded, self-revision keeps sanding until the voice goes smooth and dead. One pass catches the real problems and stops before the polish becomes the problem.
Be clear about what the gate is. It's a screen, and it catches most structure breaks and invented specifics before you see them. Your read is still the quality control that ships.
Save the Two Prompts and Run Them in This Order
Save the two prompts as two files in your prompt library. Mine live in an ads/damaging-admission/ folder, next to the other templates built the same way.
Run them in a fresh chat: teaching prompt, wait for "Ready," then the filled execution prompt. For more ads in the same session, ask for the next one with a different admission emphasis or a different proof detail leading. After three or four generations in one chat, start a new one, because the ads begin rhyming with each other instead of with the seeds.
If you're running a Claude Project per client, the teaching prompt works as project knowledge, and then each new ad only costs you the execution block.
Expect the block to take longer than the generation. That's the correct ratio. The fields are where the thinking happens, and ten careless minutes there produce exactly ten careless minutes' worth of ad.
Expect a strong first draft, most of the way to usable. The last stretch is your edit: sharpen the admission, swap the line Claude loved too much, cut a proof detail that reads vague. The draft arrives structure-true, and structure was the expensive part.
Then build outward. The second template takes half the afternoon, and the third takes an hour. Three templates from three different classics give you three distinct arguments to test against any offer, and that range is what you want on the table when the next brief arrives.
The Three Jobs the Template Can't Do for You
It can't make the framework fit every offer. The damaging admission needs a true flaw that costs something. Forced onto an offer without one, it produces the humblebrag, and readers smell "our only weakness is caring too much" instantly. The teaching prompt tells Claude to refuse those offers. Believe it when it does, and reach for a different classic instead.
It can't supply the judgment that corrects the extraction. Claude names moves plausibly, and plausible is the danger. If you can't articulate why a move works, the template encodes your uncertainty and repeats it in every ad it writes. The correction pass in Step 2 is short, but it's the step that separates your template from a generic one.
It can't finish the ad. It writes a structure-true first draft with honest facts. The line-level voice, the final cut, and the call on whether the ad is good enough to carry your name stay with you. That's the arrangement that keeps the byline worth something.
The Master’s Memo
The archives are full of structures that already proved themselves, and extraction takes one afternoon.
Take the Avis template, run it on a live offer this week, then pick your second classic yourself.
More clicks, cash, and clients,
Mark Masters
