Hey, Mark Masters here.

Last Friday afternoon I handed Claude the opening of a sales page it had written an hour earlier and asked it to rewrite it the way Joe Sugarman would have.

It came back looking exactly like him. Short sentences, one-line paragraphs, and little nudges scattered through it: "That's good news. Why is it good news?"

Then two separate Claude sessions marked every place a reader could stop reading. The Sugarman version had more than twice as many as the original.

I've quoted Sugarman twice in this newsletter. This is the issue where I find out what he was actually doing, and build the check that catches Claude getting it wrong.

In July I quoted Sugarman's slippery slide in the issue on scripting the reader's reaction. In August I quoted it again in the long copy issue, a few paragraphs above a prompt that cuts dead lines.

Both times I used his idea as a caption on the way to something else. I never looked at the place it points to, which is the gap between two sentences.

A cut pass judges sentences one at a time. It can hand you a page with nothing dead in it that a reader still puts down halfway, because momentum gets built in the joins. Claude's Sugarman impression got the look of his sentences right and the joins wrong, which makes it the best teacher I've found for what goes on there.

Joseph Sugarman started JS&A in 1971, in the basement of his family's house in Northbrook, Illinois. He sold electronics by mail. Pocket calculators came first, then digital watches, at a time when most of his readers had never held either.

He said JS&A was the first company to sell a pocket calculator by mail and the first to take credit-card orders on an 800 number. His obituaries repeated both. I haven't found anyone who checked, so I'm passing them on as his claims.

His reader was a stranger holding a magazine, looking at an expensive gadget they had never seen work. There was no salesperson, no demonstration, and no way to ask a question. If the copy let them stop, the sale was over, because nothing else in the transaction could pick them back up.

So the page had to carry them from the first line to the phone in one sitting.

He was named Direct Marketer of the Year in 1979. That was also the winter a Chicago blizzard kept most of his staff at home while his computer system failed, and orders went out late. The FTC came after JS&A under its mail-order rule and wanted a $100,000 penalty.

Sugarman answered the way he answered everything. He wrote an ad.

It ran under the headline "FTC Revolt" in The Wall Street Journal and The Washington Post, and then in dozens more papers and magazines. The Justice Department sued on the FTC's behalf in 1981, and the appeals court upheld the agency's authority in 1983.

He lost. I'm including it because a Legends issue that only prints the wins is a eulogy, and because it shows where his instinct went under pressure: a long page he trusted strangers to read to the end.

In 1986 he launched BluBlocker and took the method to television. He appeared in the infomercials himself, with strangers on the street trying the glasses on while the camera watched their faces, and the job stayed the same: keep them watching until they pick up the phone.

He spent his later decades teaching it, in books including Advertising Secrets of the Written Word and The Adweek Copywriting Handbook. He died in March 2022, at 83.

Sugarman called it the slippery slide. In his account, every element of an ad exists to get the reader to the first sentence, and the first sentence exists to get them to the second.

Look at what that asks of one sentence at the moment it ends. It has two jobs at once. It has to give the reader something, by answering what the last sentence opened or by adding something new. And it has to leave something open that only the next sentence can close.

That gives you four kinds of sentence.

The slide is a page built from the top row.

Both jobs have to happen in every sentence. A signpost does the second job without the first. "Those fixes fall short for a reason" opens a question and gives the reader nothing for the sentence they just spent. An exit does the first job without the second. Most bad joins are one or the other.

Sugarman had a name for that first kind of line. He called them seeds of curiosity, short lines at the end of a paragraph that tip the reader into the next one. By the table above, a seed is a signpost.

Whether seeds did more for a magazine reader in the 1970s than they do now, I can't tell you. On my test page they cost more than they paid, and Claude's impression of him was full of them.

I made up a product for the test, a $189 window insert called Nightpane, for renters who sleep beside a busy street. It's a clear panel that presses into the frame and traps a pocket of air against the glass.

One fresh Claude session wrote the opening 300 words of the sales page from a fact sheet. Two more, each seeing only the copy with no offer and no brief, then ran the audit printed below on it, marking every sentence as a link, an exit, a signpost or dead.

The cold draft was good. It opened on a scene at 2:40 in the morning, a motorbike pulling away from the lights outside your window, which is the move I'd have made. About six in ten of its sentences were links.

Its exits came in a cluster. Every audit I ran marked at least two of these three street noises as exits, and most marked all three:

Each is a complete little picture. The siren rises, passes and fades inside six words, so the reader has the whole event before the full stop.

Every sentence arrives finished. That's what model prose does well, and it's why the problem survives every editing pass aimed at sentence quality.

Then the impression. Asked to rewrite the draft "in the style of Joseph Sugarman's slippery slide," Claude split 23 sentences into 41, and links fell to under four in ten. It got there two ways.

It copied his punctuation and his seeds, and split every job in two. A gap is a decision about what to hold back, and nobody told the model what to hold back.

That middle row has a story behind it. My August cut pass deleted five lines from the cold draft, including "You've probably tried the usual fixes" and "Those fixes fall short for a reason."

My first version of the audit asked only whether a sentence left something open, and both of those do. So I spent an afternoon writing this issue to tell you my August prompt deletes the very sentences Sugarman built his pages from.

The rebuilt audit asks the second question as well, and both lines came back as signposts in every call but one. With them gone, the earplug sentence carried the question on its own, and links went from about six in ten to about three in four.

The cut pass was right, and my first audit was too soft to see it. That's why the version below asks two questions at every sentence.

Here's the audit, written as a Claude Skill so it fires whenever you review a draft for flow.

In Claude Code, save it as ~/.claude/skills/exit-point-audit/SKILL.md. In the Claude apps, put the file in a folder named exit-point-audit, zip the folder and upload it from the Skills section of your settings. Without Skills, paste everything from "# Exit-Point Audit" down into a fresh chat and put your copy underneath.

Four things in there are doing the work.

The audit finds the exits. The repair is the writing, and I went through three versions of a repair prompt before I trusted one.

The first rewrote the whole passage, and the exits moved down a line. It fixed the horn sentence by ending it on "the earplug meant to stop all this is on your pillow." The next two sentences, about earplugs and white noise, became the answer to that, and both auditors marked exactly those two as the page's only remaining exits.

Run twice, it scored higher and read worse. The second pass produced "It slipped onto the pillow, as earplugs do, and if you can't sleep in them at all, you've probably tried fighting noise with noise." One auditor gave that version a perfect score, the best of the whole test. The audit measures joins, and it can't hear a bad sentence.

The second version gave options, checked the sentence after each fix, and was allowed to refuse a tangled one. Its sentences were clean, but it fixed exits by bolting a clause onto a finished picture: "A siren rises, passes and fades, and nothing drowns it out." The siren still finishes, so the reader still feels finished, and the score didn't move.

The third version, below, adds one rule. Change where the sentence ends, not how much it says.

On its own, its assembled version still scored no better than the page it started from. The options were good: "A siren starts to rise as it turns into your street." "By the time somebody leans on a horn at a green light, you're reaching for the earplugs." But each street noise still stood as its own picture, and the auditors kept marking them as exits.

So I did what I keep telling you to do. I took the best pieces from all three runs, joined them, added one line of my own, and wrote the passage myself.

Each sentence starts where the last one stopped. The siren rises over the bar's voices and fades as the light turns green, and the horn sends you for the earplugs. That's the slide. The first prompt found half of it and broke the sentences after, and the third found clean pieces and never joined them.

The hand-built version had no runs of exits anywhere. Both auditors still flagged the same gap: the page opens on a motorbike and never says whether Nightpane stops one. That's a question for the client, and no rewrite can answer it.

That's where I come down. Use Claude for the audit every time. Use the repair prompt for options, write the fix yourself, and read the sentence after it before you move on.

Sugarman engineered the slide for a reader who had already picked up the magazine. A scrolling page with a notification arriving mid-paragraph is a different environment, and no audit can see a notification.

It also fits some pages badly. A pricing table, a docs page or an FAQ works better when the reader can leave the moment they have the answer. Momentum on those is an obstacle.

A structure built to make stopping difficult deserves a position, and mine is the bullet gate's rule: every gap has to be real, and the payoff has to arrive. That's what the owed ledger is for. A slope built over nothing wastes more of the reader's time than a flat page does.

The test is small, so here's what it can't carry.

Claude can copy Sugarman's sentences in seconds. The slide lives in the joins, where one sentence has to give the reader something and open the next in the same breath.

Run the audit, write the fixes yourself, and read the sentence after every fix, because that's where the exit goes when you close it.

More clicks, cash, and clients,
— Mark Masters