Hey, Mark Masters here.
Last Friday I asked Claude to write the close of two sales pages. One sold a $29 sourdough starter kit. The other sold a $3,000 pricing program for small design studios.
Four models wrote two closes each for each page, sixteen in all. I expected a pile of "Learn more" buttons.
I got buttons better than most of the pages I review.
What I didn't expect was the pattern underneath them:
At $29, every close put the price on the button and then told the reader there was no hurry.
At $3,000, most of the closes moved the price away from the button, and five of the eight made up numbers the fact sheet never gave them.
Here's the test, and the prompt that fixed most of it.
The Close Is the One Place the Page Has to Ask Out Loud
You know what a close has to carry. The ask, the price, what happens after the click, the last doubt answered, the guarantee in its real terms, and a deadline if there is one.
Everything above the close persuades. The close has to ask, and asking is the one move on the page where the writer takes a risk. A reader can say no to a sentence that asks. Nobody says no to a sentence that only describes.
That's the move I wanted to watch Claude make.
I wrote fact sheets for two products that don't exist.
Crumbwell Starter Kit | Studio Rebuild | |
|---|---|---|
Price | $29, one time | $3,000, or three payments of $1,050 |
What it is | Dried sourdough culture, a guide, a banneton liner and a 14-day email course | A 12-week small-group pricing program for 2-to-10-person design studios |
Guarantee | Free replacement if it isn't bubbling by day 7, or a full refund within 30 days | Full refund at the end of week 3 if you attend the first three calls and complete the audit |
Deadline | None | Enrollment closes January 3, or when the 12 seats fill |
Proof | 4.8 average from 1,312 reviews, plus one named review | 41 of 64 past owners reported back, median project fee up 31%, plus one named testimonial |
Each fact sheet came with a summary of the page above the close:
The headline, the opening story, the mechanism and the proof. Then came the plain request a working copywriter would type: write the close, everything after the proof section, through the button.
Four Claude models (Haiku, Sonnet, Opus and Fable) each ran it twice per page, every session clean with nothing else loaded. A separate session scored every close against a fixed rubric and had to quote the line behind each call. I checked its calls myself.
I Expected "Learn More." The Buttons Were Better Than That.
Credit first, because Claude earned it.
Fifteen of the sixteen buttons named the thing the reader was buying. "Send Me the Starter — $29." "Claim My Seat in the January Cohort." "Get My Starter Kit: $29." Only one was stock, a plain "Reserve my seat."
Every close answered a specific doubt before it ended. All but one stated the guarantee in its exact terms.
And eight of the twelve runs on the studio page caught a problem in my own fact sheet. Three payments of $1,050 come to $3,150, not $3,000. One of them flagged it as "a visible discrepancy" that "can cost trust at the point of purchase."
That's a better catch than I made when I wrote the sheet.
So the craft is there. The problem sits in one or two sentences near the button, and which sentences depends on the price.
At $29, Every Close Told the Buyer There Was No Hurry
My fact sheet said: "Deadline: none. The kit is always available at this price."
That line was for the writer, so nobody would invent a countdown.
All eight $29 closes passed it straight to the reader.
"So take the time you need."
"Nobody's rushing you."
"There's no countdown on this page and no price going up at midnight."
"The kit is always $29, and it'll be $29 next month."
"Order whenever you're ready to start."
Read those from the buyer's chair. You've scrolled to the bottom, you're mostly sold, and the page tells you it's fine to leave.
C'mon. You'd never write that line for a client.
The honest instinct was right. A fake countdown on a product that never goes off sale is a lie, and readers spot it. But "no deadline" is a fact the copy should keep to itself. The close's job at $29 is a true reason to start today, and the fact sheet had one: the culture is ready to bake with in about five days.
Opus and Fable both found one. Fable wrote, "The only thing that changes if you wait is which Saturday you pull your first real loaf out of the oven." Opus asked "how many more jars of flour and water you want to pour down the sink."
Both lines came right after telling the reader there was no countdown. Delete the first sentence and keep the second, and you have a close.
At $3,000, Five of Eight Closes Moved the Price Off the Button
The studio closes ran longer, between about 420 and 970 words, and they read well. Then I looked at where the price was.
On the $29 page the price was on the button eight times out of eight. On the $3,000 page it was on the button three times out of eight. In the other five it sat higher up the page and came back, if at all, as small print under the button.
Four of the eight had no sentence asking the reader to buy at all. Two more ended on a choice or an "if you're ready" instead: "You can keep treating those as two problems, or you can spend twelve weeks fixing the one they actually are." Then the button.
The $29 close says the price out loud. The $3,000 close mostly whispers it.
That's backwards. At $29 the price is barely a decision. At $3,000 the price is the decision, and the reader needs it right next to the thing they're about to click, with the payment option and the guarantee beside it. Otherwise the price is something they meet in small print, or at checkout.
The split followed the model. Haiku and Sonnet whispered it all four times. Opus and Fable put it on the button three times out of four, as in "Claim My Seat: $3,000" and "Enroll in Studio Rebuild — $3,000 or 3 × $1,050."
Peggy's upsell teardown found that none of Claude's cold drafts closed on the purchase the buyer had just made. It's the same reluctance to ask, one page further down the funnel.
Where the Facts Ran Out, the Close Wrote Some
The close is where a page needs weight. When the fact sheet didn't supply it, the models made it up in two ways.
The first was invented customers and invented scarcity. Six of the sixteen cold closes said something about past buyers, or about seats running out, that the fact sheet never said. Every one of the six came from Opus or Fable.
What the close said | What the fact sheet said |
|---|---|
"Most people who buy this kit have." (killed a starter before) | Nothing about who buys it |
"Dana killed three starters before this one." | Dana's review says her starter doubled on day six |
"The replacement requests are rare." | Nothing about replacement requests |
"Past cohorts have filled before the deadline." | Nothing about past cohorts filling |
"The next cohort after this one has not been scheduled." | Nothing about future cohorts |
Every one of those is a line you'd love to have. Every one is a line you'd have to get from the client before it goes on a page with their name on it.
The second was payback math on a number nobody had given them. Five of the eight studio closes did it. "If your average project is $10,000, a 31% increase is $3,100 on a single project." Four picked $10,000 and one picked $5,000.
Opus didn't, either time. One run handed the arithmetic back to the reader: "Take your average project fee and multiply it by 0.31."
That's the right instinct with the right number. A $3,000 close should show the money coming back. The number has to come from the client or the reader, and the honest version of that line starts as a question in your brief: what's your average project fee?
I can describe these patterns from the outside, and I can't prove what's going on inside the model. My read is that the close is where the page most needs weight, so the model reaches for the strongest material it can find. When the facts run out, it writes some.
Peggy caught the same move last week with "make it more specific". Where the brief was thin, the padding showed up at the same confidence as the facts.
Write the Ask First, Then the Button, Then the Close
Here's the prompt I ran against the plain request, with the same fact sheets and the same four models, one run each. It's printed as tested, with the two paste-in fields added at the bottom.
You're writing the close of a sales page: everything after the proof section, through the button. The page above it has already made the case. The close has one job, which is to ask for the order plainly and make yes the easy answer for a reader who is already mostly persuaded.
Work in this order.
1. Before writing any copy, answer in one line each:
(a) The exact action you are asking for, the price the reader will pay, and what happens in the first minute after they click.
(b) What saying no costs the reader, taken from the problem the page has already described and the fact sheet. Do not invent a deadline or a loss the facts don't support.
(c) The one doubt most likely to stop this reader at this price, and the fact in the fact sheet that answers it.
2. Write the ask: the single sentence directly above the button. It names the action and the price, in the reader's terms. Write it before anything else in the close. Do not soften it with phrases like "if you're ready", "whenever you're ready", "no pressure" or "feel free".
3. Write the button text from the ask: its verb and its object, in the reader's voice. Never use "Learn more", "Submit", "Click here", "Get started" or "Sign up".
4. Build the rest of the close around the ask, scaled to the price.
- Under $100: keep it short. Restate what they get, remove the last bit of friction (shipping, refund, what happens next), and ask once.
- Over $1,000: answer the doubt from 1(c) before the ask, show what staying put costs from 1(b), state the risk reversal in its exact terms, state any real deadline once, and give the payment option if there is one.
Ask once, with one button. If the page needs a second button, it uses the same text.
Use only facts from the fact sheet. Add no number, testimonial, scarcity or deadline that isn't there. Add a P.S. only if it carries a fact the close didn't.
Return your answers to step 1, then the finished close with the button text in [square brackets].
FACT SHEET:
[Paste the offer: what it is, the price and any payment plan, the guarantee in its exact terms, any real deadline or seat limit, the proof you can actually use, and what happens after checkout.]
PAGE SO FAR:
[Summarize the sections above the close: headline, opening, mechanism, proof. The close can only pay off what the page has already set up.]Five things in there are doing the work.
Step 1 makes it say the price before it writes any copy. In every guided run, the price it wrote down in step 1(a) came straight through into the ask.
The ask gets written first, and the softeners are named. I listed the actual phrases rather than asking for a confident tone, because the actual phrases are what turned up in the cold runs.
The button is built from the ask. That keeps the button and the sentence above it saying the same thing, so the click confirms a decision the reader just read.
The close changes shape with the price. Under $100 it gets out of the way. Over $1,000 it does the heavy lifting before the ask. One shape can't serve both, and the cold runs showed what happens when the model picks its own.
The facts rule names numbers, testimonials, scarcity and deadlines. Those cover everything the cold closes made up.
What the Prompt Fixed, and the Two Things It Broke
Plain request (16) | Close prompt (8) | |
|---|---|---|
A plain ask with the price, directly above the button | 0 | 8 |
Told the reader there was no hurry | 8 | 1 |
Invented claims about past buyers, or invented scarcity | 6 | 0 |
Payback math on a figure nobody supplied | 5 | 0 |
The $3,000 closes got the most out of it. Every guided studio close put the price and the payment option in the sentence directly above the button. Opus wrote the best line of the whole test there, under the heading "Here's what saying no keeps": "Every project you price between now and January gets priced the way the last ten were."
That's what staying put costs, in the buyer's own calendar, with nothing invented.
Two things broke.
The $29 closes shrank to less than half their length, from 316 to 514 words down to 143 to 194. Some of that was fat. Haiku's guided version read like a packing list, though, and the warmth went out with the fat.
The four $29 asks came back in the same shape.
"Order your Crumbwell Starter Kit for $29, one-time, with free US shipping."
"Get the Crumbwell Sourdough Starter Kit for $29, one time, with free US shipping."
"Order your Crumbwell Sourdough Starter Kit for $29, shipped free, and skip the two weeks where starters die."
I asked for a plain ask, and plain converged. Peggy saw the same thing in her upsell test: the prompt fixed the close, and every session wrote the same close.
One more thing. Fable's P.S. put the no-hurry line straight back, after the prompt had limited the P.S. to new facts: "There's no deadline on this. The kit is always $29." Read the P.S. first. It's where the model goes when the close is finished and it still wants to say something.
The Line Before the Ask Is Still Yours
Here's where I come down.
The prompt makes sure the ask exists, names the price and sits directly above a button that says the same thing. That's most of a close, and it's the part Claude skipped on its own.
The sentence just before the ask is where the reason to act today goes. The best ones in the test only turned up when the fact sheet handed the model something true to say:
"The only thing that changes if you wait is which Saturday you pull your first real loaf out of the oven."
"Every project you price between now and January gets priced the way the last ten were."
"Take your average project fee and multiply it by 0.31."
Write that sentence yourself. Then check the facts under it, and check whether your client has an average project fee you can put in place of the one Claude would invent.
Where This Test Is Thin, and What It Can't Tell You
Two invented offers, one consumer and one B2B. A $300 offer, or a close that books a sales call instead of a checkout, may behave differently.
Sixteen cold closes and eight guided, with one guided run per model. That's enough to see a pattern this consistent and too few to put a rate on it.
Nobody bought anything. The scoring counts what's on the page. I'd bet on the ask above the button. I wouldn't bet on any particular conversion figure.
The split between models is real but thin. Haiku and Sonnet hid the $3,000 price, and Opus and Fable invented the customers. Two cold runs per model per page is enough to see that and too few to call it a property of the models.
Every scorer was Claude. It counted strictly, flagging lines like "Most people have." as inventions, so I've only reported the categories I checked by hand and would defend.
The Master’s Memo
Claude will write you a good button at any price.
It won't reliably put $3,000 next to it, and at $29 it will tell the buyer there's no rush.
Write the ask first, with the price in it. Keep the line before it for yourself.
More clicks, cash, and clients,
— Mark Masters
