Skip to content
andydataguy
Wiki

Voice Compiler: Rules from approved copy. Approved copy can teach an AI writer which changes to preserve.

AI & TECHNICAL · Deep Dive~77 min read · 18,282 words
A bolt of amber cloth on a cutting table feeds three lengths forward to three headless dress forms on one stone plinth: a long overcoat on the left, a work jacket in the middle and a short vest on the right. A violet stone wall stands behind all three.
One piece of content goes out on three channels. The cloth stays the same, each cut fits the people reading on that channel, and one wall of shared minimums stands behind all three.

Many job posts I've read from people who buy writing at volume describe the same week. A draft comes back sounding like nobody in particular. The buyer fixes a sentence, and the next draft carries the same sentence again. By the third time they're fixing the same sentence, they're paying for a writer and doing the writing.

On a call about a batch of social posts, the owner of an equipment dealership told me he can't judge grammar and wasn't going to try. He sells construction positioning gear to the people who run excavators and graders, and the posts were for them. He said he'd read each one to make sure it made sense and that the posts "have the right vibe and they sound right." I'd worried out loud that one product description might be talking nonsense about how the blade works. He read it and said, "It's talking correctly."

He'd split the job in two. The half he handed off, grammar, is work that written rules and a careful reader can carry. The half he kept is whether a post makes sense to an operator and talks like someone he'd trust. His ear decides that, or the ear of someone who sits in a cab all day.

Before splitting anything, I wanted to know where faults enter a piece of writing, so I traced them through work I'd already shipped. I write articles for an AI-tools publisher, and for ten of them the first draft and the text I delivered both sit on disk with nothing between them but my own pass. My pass changed about a quarter of the first draft's words: a mean of 27%, from 12% on the lightest article to 66% on the heaviest. Across 13 published pages, the publisher's team changed about 8% of what I delivered on the way to the live site.

Two separate reviews then read those texts line by line against a whole book on sentence craft, and a script traced each sentence fault they logged back to the earliest version that held its words. Between 73% and 86% of the faults it could trace entered in my edits. Most of them sat in sentences I'd added or made longer.

Three sheets side by side: the first is clean, the middle one is crowded with amber pencil marks and the third carries only a few. Labels name them the first draft, the text I delivered and the published page. My pass changed about a quarter of the words and is where 73% to 86% of the traceable faults entered; the publishing pass changed about 8% more and added a dozen errors.

FIRST DRAFT
MY PASS
About a quarter of the words changed: a mean of 27% across 10 articles. Between 73% and 86% of the faults that could be traced entered here.
THE TEXT I DELIVERED
THE PUBLISHING PASS
About 8% more changed, across 13 pages. It fixed about 1 flagged problem in 10 and added a dozen new errors.
THE PUBLISHED PAGE
Three versions of each article: a first draft, the text I delivered, and the page the publisher put live. My pass changed about a quarter of the words and is where most of the faults that could be traced entered; the publishing pass changed about 8% more and added errors of its own.

The trace locates the faults that survived to the client. It can't grade the first drafts: nobody ever scored them, and I'd already cut their weakest lines, so the first-draft words that survived are the ones I chose to keep. My pass also did the part only I could do. One first draft closed a story with "That proposal won." I changed it to "The proposal didn't win that particular client," because it didn't. Another had me building chatbots on early platforms, with no count. I'd built dozens, so the sentence says dozens. The hand that puts in a story's truth and its lived detail is the same hand that drops a clause on the wrong noun.

The publisher's team had its own pattern. On the way to the live site it fixed about one flagged problem in ten, and it added a dozen new errors, among them a figure that contradicted the body of its own article.

So the work runs in layers, and the sentence checks run again after every hand that touches the text, mine and the publisher's included. A draft settles the structure: what each part says and whether it earns its place, which is the job I've called a content compiler. A line edit makes the sentences easy and pleasant to read, the way a publisher's line editor works through a manuscript once its structure is settled. A copywriting pass makes the parts that have to persuade do that job. Under all three sits the part that makes the result sound like one business talking to its own customers on one channel. I call it a voice compiler. It keeps a written profile of how that client talks, and every layer reads from it. Under the profile sits a short list of minimums every sentence meets, whoever the client is: no em dashes, no claim without someone who saw it, no headline that stops before its thought does. At the end there's a reader's ear that no model stands in for.

A stack. The draft at the bottom settles structure, the line edit above it makes sentences good, and the copywriting pass on top makes the persuading parts great. A voice compiler band runs beside all three, the floors sit underneath, a reader's ear caps the stack, and an arrow sends copywriting changes back through the line edit. Each part links to its section.

THE STACK, AND WHERE EACH PART IS TAUGHT
THE EAR
A reader from the audience decides whether it sounds right.
GO TO SECTION
THE COPYWRITING PASS
GREAT
Makes the parts that persuade do it.
GO TO SECTION
THE LINE EDIT
GOOD
Makes every sentence easy and pleasant to read.
GO TO SECTION
THE LINE EDIT AGAIN
on every sentence the copywriting pass touched, and after every hand, mine and the publisher's included
GO TO SECTION
THE DRAFT
STRUCTURE
Settles what each part says and whether it earns its place.
GO TO SECTION
THE VOICE COMPILER
One client's voice profile. Every layer reads it.
One client's voice profile. Every layer above reads it.
GO TO SECTION
THE FLOORS
Minimums every sentence meets, whoever the client is.
GO TO SECTION
Three layers, each adding one quality: structure, then good, then great. One client's voice profile runs through all three, the floors sit under every sentence, and a reader from the audience has the last word.

I've been running all of it by hand on client work, with a person or a second model checking every operation, to find out whether the quality is reachable before anyone builds the software. On one round of the dealer's posts, a reviewer model that hadn't written a word of them graded 33 rewritten captions and flagged 10 as major problems. After the next revision, it flagged none. Some of the quality is reachable now. The rest needs the client's ear, and my method treats that ear as data to collect.

Here's the test the split has to pass. The dealer's posts go out on LinkedIn, Facebook and Instagram. Most of a post's content can be shared across all three. The voice on each channel can't. If the three versions read alike, the voice compiler did nothing, and an operator who follows the dealer on two of those channels will hear it.

AI copy sounds like everybody for three separate reasons

Reading alike is the default, because a language model writes toward the average. Ask it for a post about grading blades and it returns the middle of every post about grading blades it has absorbed: polished, balanced and safe. The writing standard I hold my drafts to defines bad machine writing as "average generalization on a concept, delivered in a way that helps nobody learn." The result has a consultant's cadence, a conference talk's vocabulary, and no particular person underneath.

When a buyer tells a vendor a draft "sounds like AI," three separate failures are hiding under that phrase, and each needs its own fix.

The first failure is content. Parts of the piece do no job for the reader: an intro that restates the headline, a paragraph explaining a problem the reader already lives with, a summary that drops the condition the body was careful to state. The second is voice. Every sentence is fine and none of them belongs to anyone, so the post could run on a competitor's page without a word changed. The third is the tells, the surface habits readers have learned to spot: em dashes, the same dozen inflated verbs, a contrast that knocks down a claim nobody made, a line that tells the reader what they feel.

Fixing one leaves the other two where they were. I've seen drafts scrubbed of every em dash and every inflated verb that still had no point, and drafts with a sharp point that read like a brochure. A banned-word list can only reach the third failure.

Three stacked bands. Content asks whether each part does a job for the reader, and the draft fixes it. Voice asks whose mouth the words come out of, and the voice compiler fixes it. The tells, at the base, are habits readers spot, fixed by the floors, and grammar is a thin line inside that base band.

THREE WAYS A DRAFT SOUNDS LIKE AI
CONTENT
Does each part do a job for the reader?
An intro that restates the headline.
FIXED BY THE DRAFT
VOICE
Whose mouth is it coming out of?
Fine sentences that could run on any competitor's page.
FIXED BY THE VOICE COMPILER
THE TELLS
Would a reader spot the habits?
Em dashes, inflated verbs, an invented quote.
FIXED BY THE FLOORS
GRAMMAR

Fixing one leaves the other two where they were.

When a draft sounds like AI, three separate things can be wrong: parts with no job, a voice that belongs to nobody, and the tells. Grammar is one thin line inside the smallest of them.

One summer draft for the dealer, which never shipped, failed all three at once. It opened, "Ask a survey lead how the first morning of the season went and the best answer is a quiet one." It went on to quote a crew member nobody had interviewed, and it closed on a tidy line about plain talk beating polish. The content failed first: the post never said what a reader gets, and the owner sent it back saying he couldn't tell what it was about. The voice failed with it. On drafts like that one, I wrote: "Who is the human? Speak to them. This is not speaking to anybody." The tells were the invented quote and the aphorism for a closer. Cleaning up the tells alone would have left a grammatical post about nothing.

In the one place I've measured it carefully, the tells were also the smallest part. Fifteen articles I'd delivered to the AI-tools publisher went through a line-by-line review by a separate model, which I'll call the first review. It had read only the contents and first chapter of that book on sentence craft, and it used the other chapter titles as lenses. Fourteen of the fifteen articles went on to an approved final. The first review logged 1,368 notes. The biggest category was claim logic and evidence fit, with 334 notes spread across all fifteen articles. Technical and product precision came next with 297, then sources and attribution with 217. Grammar drew 28 notes. Mechanics, the commas and capitals and hyphens, drew 30.

Bars for the first review's notes by category across fifteen articles I delivered: claim logic and evidence fit 334, technical and product precision 297, sources and attribution 217, word choice 100, voice and stance 74, concision 74, mechanics 30, grammar 28.

THE FIRST REVIEW: 1,368 NOTES ON 15 ARTICLES I DELIVERED
Claim logic and evidence fit THE CLAIMS
334
Technical and product precision THE CLAIMS
297
Sources and attribution THE CLAIMS
217
Word choice and terminology
100
Voice and stance
74
Concision and repeated meaning
74
Mechanics THE COMMAS
30
Grammar THE COMMAS
28
THE CLAIMS
THE COMMAS

Selected categories. These count review notes, not error rates: one problem can earn several notes. This review read the book's contents and its first chapter.

A line-by-line review of fifteen articles I'd delivered, made with only the book's contents and first chapter in hand, logged 1,368 notes. Claims drew about twelve notes for every one about grammar.

The reviewer was careful to say those are counts of review notes and not error rates, since one problem can surface in a summary, a table and an FAQ and earn three notes. Even allowing for that, claims drew about twelve notes for every one about grammar. Add the three biggest categories and 848 of the 1,368 notes, about 62%, sit in claim logic, technical precision and sources. The book's sentence-craft checks don't replace a separate check of claims, technical facts and sources. That split is why the claim checks belong to the draft, before anyone edits a sentence for how it reads, and why the sentence work belongs to a line edit that never touches a claim.

I also tried having one model check another model's drafts, and measured what that bought. Three drafts had passed every string check I'd built at the time: no em dashes, no banned words, clean on every lint rule. Two model reviewers, each blind to the other, scored them 77 and 88 out of 100. I scored the same three around 30 and called them obviously machine-written.

The gap between those scores made the judges unsuitable for that decision. Shared writing conventions are one possible explanation: a model can reward habits the audience dislikes. Those three drafts don't establish why the judges disagreed with me, or how every model reviewer behaves. They do show why I need a more specific question and a reference set before relying on a score.

The work has to split the way the failures split, and editors were dividing it along those lines long before anyone had a model to worry about. A manuscript at a publishing house goes through several readers, and each one reads a different unit of the text. One reads the whole book and asks whether chapter four should exist. Another works paragraph by paragraph on how the prose moves. A third fixes commas, spelling and cross-references, and a last one checks the typeset pages. The comma reader isn't the one who tells the author the middle third of the book is in the wrong order.

Editors Canada writes the first of those jobs into its professional standards. Structural editing, in its words, "is assessing and shaping the overall organization and content of the material to optimize it for the intended audience, medium and purpose." The standard tells the editor to make deletions "to remove repetitive, irrelevant or otherwise superfluous text or other elements" and additions "to fill gaps in content or strengthen transitions between sections."

The Editorial Freelancers Association defines the levels below it: line editors work "at the sentence or paragraph level," copyediting means "correcting spelling, grammar, usage, and punctuation," and proofreaders check the finished pages.

Four rungs of a ladder. A structural edit reads the whole piece and maps to the draft. A line edit reads paragraphs and sentences and keeps its name. A copyedit maps to the floors, and a proofread maps to a check of the live page. Beside the ladder, a separate card holds copywriting, its own layer.

WHAT EACH READER READS, AND WHERE IT SITS IN THE STACK
STRUCTURAL EDIT
Reads the whole piece and its parts
THE DRAFT
LINE EDIT
Reads paragraphs and sentences
THE LINE EDIT
COPYEDIT
Reads sentences and words
THE FLOORS
PROOFREAD
Reads the final page
LIVE-PAGE CHECK
after it ships
COPYWRITING
From direct response: its own layer, judged by whether the reader acts.

Off the ladder, the voice compiler holds one client's voice profile, and every rung reads it.

Publishing houses already split editing by the unit each reader works on, from the whole book down to the final page. Each rung maps to one layer of my stack, and copywriting, which comes from direct response, sits beside the ladder as a layer of its own.

When I describe what I want from a content system, I end up asking a structural editor's questions. Does this scene add value? Does this character have the impact it's looking for? Is the overall arc worth making? The draft layer puts those questions to a blog post's third section or a landing page's proof block, every part of every piece, and writes each answer down.

The ladder maps onto the stack one rung at a time. The structural edit is the draft, the job I've called a content compiler. The line edit keeps its name and its unit, the sentence. The copyedit becomes the floors, the minimums every sentence meets whoever the client is. The proofread becomes a read of the live page after it ships, and the publisher's edits to my articles show why that read can't be skipped.

Two parts of the stack sit off the ladder. Copywriting comes from direct response, where a line is judged by whether the reader acts, and it gets its own layer because "easy to read" and "moves the reader" are separate tests. A headline can be perfectly clear and give nobody a reason to read the line under it. The voice compiler sits off the ladder too. It holds the profile of one client's voice, and every rung reads that profile, from the structural questions down to the proofread.

The order follows from the units. The draft goes first and settles what each part says. The line edit comes next and moves no paragraph; a structural problem it finds goes back to the author as a question. The copywriting pass then works on the parts meant to persuade, the headline, the opening, the proof, the offer and the ask. Whatever it rewrites goes back through the line edit, because new slips get in where a line is rewritten, as the record of my edits measures. Draft, line edit, copywriting pass, then the line edit again on every sentence the copywriting pass touched.

Each layer adds one quality the layer before it can't. The draft gives the piece its structure: every part has a job, and every claim is the size of its evidence. The line edit makes it good, sentences a reader moves through without stalling or going back. The copywriting pass makes the persuading parts great, so the headline earns the next line and the ask gives the reader a reason to act. Skip the draft and the later layers polish a part that shouldn't exist. Skip the line edit and the persuasion lands on sentences that trip the reader. Skip the copywriting pass and the reader finishes a clear, accurate piece and does nothing about it.

A rough-sawn plank, the same plank planed smooth and a finished tabletop with an amber inlay stand on one stone plinth, sharing one grain and one knot. Labels name them the draft, the line edit and the copywriting pass; the grain is the client's voice profile and the stone is the floors.

THE DRAFT
Structure: every part has a job, every claim its evidence.
THE LINE EDIT
Good: sentences a reader moves through without stalling.
THE COPYWRITING PASS
Great: the parts that persuade do it.
THE GRAIN
The client's voice profile, the same from rough plank to finished top.
THE STONE
The floors, under every stage.
Whatever the copywriting pass rewrites goes back through the line edit.
Writing goes from rough to good to great in three layers, each adding one quality the layer before it can't. The grain running through all three is the client's voice, and the stone under them is the floors.

Every layer after the draft works on sentences, and no sentence can rescue a part that shouldn't be there.

The draft settles what each part says and whether it earns its place

My earlier wiki entry on the content compiler laid out the architecture for this first layer, and the architecture holds. A piece of content is a tree of typed parts instead of one long string: a hook, a section, a claim, a citation, a call to action. Each part is produced and checked on its own, so a revision to one paragraph re-renders that paragraph and serves the rest from cache, and one content tree can be emitted to many channels the way a compiler emits one program to many machines. The reference build shows it in practice: each paragraph is stored as its own row, and a paragraph that fails a check gets regenerated alone while every approved paragraph stays exactly as it was.

That entry's checks confirmed that a part existed, that it was formatted, that its citation resolved and that it obeyed the style rules. A check that confirms a part exists says nothing about whether it does a job. Running the operations by hand on client work showed me what the draft layer is for: deciding whether each part earns its place.

So the draft starts by breaking a piece into the parts its type calls for. A website breaks into pages, each page into sections, and each section into elements: a headline, a label, a button, a caption, a line of alt text. An article breaks into a headline and deck, an opening, body sections that each carry a claim with its evidence and example, a close, and the short forms a skimmer reads (summary, FAQ answers, takeaways, the search description). A social post breaks into four parts: the hook, the point, the proof and the ask.

Three trees of parts. A website splits into pages, sections and elements. An article splits into headline and deck, opening, body sections, close and short forms. A social post splits into hook, point, proof and ask. Each part carries a one-line job.

WEBSITE
  • Pages: one per visitor question, such as a service page
    • Sections, such as the proof block: makes the claim above it believable to someone who's never heard of the business
      • Elements: headline, label, button, caption, alt text
ARTICLE
  • Headline and deck: one complete thought a stranger could repeat
  • Opening: puts the reader inside the problem before the method arrives
  • Body sections: each carries a claim with its evidence and example
  • Close: tells the reader what to do next
  • Short forms (summary, FAQ answers, takeaways, search description): say nothing the body doesn't already say
SOCIAL POST
  • Hook: names the situation the reader is in
  • Point: one benefit the viewer gets
  • Proof: the one sourced fact that makes the point believable
  • Ask: what the viewer should do next

A part whose job can't be stated in a sentence is a cut candidate.

Every content type breaks into parts, and every part gets a job written as what it does for this reader at this point. A part whose job you can't state in a sentence is the first thing to cut.

Each part then gets a job, written as what it does for this reader at this point. A service page's proof block exists to make the claim above it believable to someone who's never heard of the business. An article's opening exists to put the reader inside the problem before the method arrives. A part whose job can't be stated in a sentence is a cut candidate, and the structural editor's moves follow from there: cut what's repetitive or irrelevant, add what fills a gap, move what's out of order, fix what does its job badly.

Each part also carries a rule about how much it may make up, and the rule depends on its type. A part that asserts something about the world traces to a source, or it stays a visible gap in the draft until someone fills it. A part that condenses the body, like a summary or an FAQ answer, may say nothing the body doesn't already say. A part the writer composes, like an analogy, a hook or a transition, has to teach or move the reader along, and any fact tucked inside it follows the sourcing rule. My standard puts the line in one sentence: you may invent a metaphor, and you may never invent a story, a figure, a quote, a source or a specific detail.

What comes out of these steps is a content map: every part with its type, its job, a verdict (keep, cut, add, move, or a question for the author) and the evidence behind any claim it makes. The map is what the line edit and the copywriting pass receive. Neither of them ever receives a loose draft.

Order is often a content decision. An audit of one batch of the dealer's queued posts found that 23 of 34 distinct first lines answered a question before the post had raised it. One draft post opened, "You can get help on a machine without anyone driving out to the job." It's a clear sentence, and it answers a question the reader hasn't been asked yet. My rewrite opened on the situation: "A machine failing outside of town can mean hours or days before support arrives." The old opener survived as the second line, almost word for word. Nothing about the wording was wrong. The parts were in the wrong order, and the first part had no job until the trouble had been named.

Before: a draft post opens with "You can get help on a machine without anyone driving out to the job," stamped move. After: a new first line names the trouble, stamped add, and the old opener follows as the second line, almost word for word.

ONE POST, JUDGED PART BY PART
BEFORE
MOVE
You can get help on a machine without anyone driving out to the job.
Answers a question the reader hasn't been asked yet.
AFTER
ADD
A machine failing outside of town can mean hours or days before support arrives.
Names the trouble first.
MOVED
You can get help without anyone having to drive out.
The old opener, almost word for word, now answers the line above.
VERDICTSKEEPCUTADDMOVEASK THE AUTHOR
A draft post answered a question before the reader knew there was a problem. The rewrite opens on a machine failing outside town and keeps the old opener, almost word for word, as its second line.

Next comes a check of claims against their evidence, and it runs before anyone touches the voice or a sentence. A capability can't be stated as a guarantee. A single experience can't be stated as a law. A method that worked once can't become the best method. A sentence that claims too much is wrong in every voice, and polishing it only makes the overclaim more persuasive.

The first review of the text I delivered found those moves again and again. One article told readers a connection would let an assistant "safely reach" their real files, apps and data. The revision dropped one word, "safely," because the sentence presented security as a built-in property of the connection. Another promised checks and review passes "to ensure quality control," and the revision changed "ensure" to "support," because the section that followed described review steps and nothing that guaranteed an outcome. That sentence wasn't in the first draft. It entered in my own pass.

One article ended a personal story about relying on a single AI assistant with a rule for everybody: it was "imperative" that people keep an alternative to that assistant. The revision kept the story at its own size: "I realized I needed a backup." A line saying the publisher's product was "rising to become the best solution" in its category came down to "That's where [the product] fits," which gives the product its job in the reader's workflow and stops. And a line saying prompt quality "directly determines" output quality became "helps shape," because a later section of the same article explained that model choice matters as well. The article had contradicted its sole-cause claim a few scrolls down.

Three before-and-after cards. "To ensure quality control" becomes "to support quality control." "It's imperative for people to always have alternatives" becomes "I realized I needed a backup." "Rising to become the best solution" becomes "That's where the product fits."

GUARANTEE TO CAPABILITY
You can add guardrails and evaluation loops to ensure quality control.
You can add checks and review passes to support quality control.
ONE STORY TO ONE PERSON'S LESSON
That's when I understood that it's imperative for people to always have alternatives to [the assistant].
That's when I realized I needed a backup to [the assistant].
BOAST TO JOB
rising to become the best solution
That's where [the product] fits.

Every fix sizes the claim to the evidence under it. None of them touched grammar.

Three fixes from the first review of text I delivered: a guarantee cut back to a capability, one person's story kept to one person, and a boast replaced by the product's job. None of them touched grammar.

None of those fixes touched grammar, and the sentence-craft book has no chapter for them. The full-book review of the text I delivered found no chapter that governs claims, so it sent each claim item to a fact check as a question for the author instead of making an edit. The line edit works the same way: when a better sentence would need a different claim, it writes a query and leaves the claim alone. The check also cuts the other way. A strong claim stays strong when its evidence supports it, and the first review asked for exactly that: keep the confident claims the evidence carries. Where a guarantee overreached, the fix names the condition under which the claim holds, instead of burying it under hedges.

Then the short forms get read by themselves. Headline, summary, FAQ answers, captions, the call to action, the search description: read only those, write down what a skimmer would take away, and compare it with what the body says. When the two disagree, the short form is wrong, because the body is where the writer had room to state the conditions.

The first review shows how a condition goes missing. One passage in my delivered text said grounding can't make an AI incapable of being wrong, and then ended, "It makes it checkable." In its place the reviewer wrote: "When the system exposes its sources, it makes the answer easier to check." The memorable version is the one that gets lifted into a takeaway or a social card, and it's the version missing the condition that makes it true. The publisher's edits after delivery did the same thing on a different page: they dropped a qualifier that the summary and the body both kept, and the takeaway they produced claimed more than either of them.

Two versions of one passage. The first ends "It makes it checkable," marked as the line a takeaway box would lift. The revision ends "When the system exposes its sources, it makes the answer easier to check," with the restored condition marked in amber.

THE MEMORABLE VERSION

Grounding does not make an AI incapable of being wrong. It makes it checkable.

WHAT A TAKEAWAY BOX LIFTS
THE REVISION

Grounding does not make an AI incapable of being wrong. When the system exposes its sources, it makes the answer easier to check.

THE CONDITION, PUT BACK

A skimmer who reads only the first version believes grounding makes answers checkable by itself.

The memorable line is the one that gets lifted into a takeaway or a social card, and it lost the condition that made it true. The revision keeps the claim and puts the condition back.

Structural faults survive approval, and no sentence pass reaches them. The full-book review of the text I delivered found one article that gave its definition five times, another that left a duplicated paragraph with the same figure twice in four lines, and a third that spent about 500 words on history and statistics before its first instruction. Every sentence in those passages could have been polished and the reader would still have waded through them.

Length brings a structural fault of its own. Measured on five real documents, average sentence length fell from front to back in four. The back half of a long draft gets shorter sentences, more bullets and thinner development, and the writer doesn't notice it happening. So a long piece gets a skeleton first, with a word target for every section, and is filled one section at a time, each section checked against its target before the next begins.

The social post is where all of this gets small enough to see at once. Before a word of a post gets written, it gets a four-line brief. The first line names the audience. The second is the point, one sentence from the viewer's side saying what they should take away. The third is the proof, the one sourced fact that makes the point believable. The fourth is the ask, what the viewer should do next. Advertising has long called the second line the single-minded proposition: the one thing the viewer should walk away with, decided before any copy exists. My rule for that line is blunt: "If you can't answer that question, you have no business writing anything else."

The point has a second test, which I added after one draft passed the first. A point can be clear, sourced and correctly understood by a cold reader, and still be worthless to the viewer. One draft's point was that the dealer adds machine control to machines built without it. That's the seller's business model. No operator spends his evening worried about which of his machines shipped without machine control. The point has to be a benefit the viewer gets, like cutting to grade on the first pass, and a point about the company, its market position or its sales process fails before anyone judges a single word.

Every defect the draft layer finds goes back to the step that owns it, and gets fixed there. A structural hole goes back to the outline. A claim with no source goes back to sourcing, where it either finds a source or gets cut. A short form that drifted gets rewritten from the body's most careful sentence. The temptation with any finished-looking draft is to patch the sentence where the problem surfaced. I've watched that patch hold for one draft and fail on the next, because the outline or the brief that produced the defect is still sitting there, instructing the next draft to make it again.

A two-storey house cut open. Downstairs, a grey bucket and mop sit under a drip from a stained ceiling. Upstairs, off to one side, a loose pipe joint leaks along a beam toward the stain, and a wrench rests on the glowing joint.
A defect shows up in a sentence and starts in the outline, the brief or the sourcing. Patch the sentence and the next draft drips again; fix the step that made it and the leak stops.

The first review closed on five questions a finished article should let its reader answer. It wrote: "What is the useful idea? What actually happens? Under what conditions is the claim true? What should I do next? Why should I trust this account?" They make a fair acceptance test for the draft layer. A piece that can't answer one of them has a structural problem, and any buyer can put the five to whatever a vendor sends back.

The next hand to touch a settled draft changes more than the sentences it set out to fix. What happens when a person edits it, the expert included, is a question I can answer with numbers from my work.

Most of the traceable faults that reached the client entered in my own pass

I can answer that question because my client work keeps its versions. For the articles I write for the AI-tools publisher, three texts exist wherever the files survived: the first draft, the text I delivered after my pass, and the final, the page the publisher put live. Fourteen of the fifteen reached an approved final. Approval meant the publisher was happy to publish. It didn't mean the articles were finished. A script compares the three texts and credits every word to the version that first held it. The full-book review I'd been learning from had read my delivered text for 14 of the 15 articles, so its lessons describe my writing more than they describe the first drafts.

Each hand changed a different amount. Across the ten articles where the first draft and my delivered text both sit on disk with only my pass between them, my pass changed about a quarter of the words: a mean of 27%, from 12% on the lightest article to 66% on the heaviest. Two articles got a later pass from me after the publisher's comments, and that pass changed 7% and 8%. On one of them the publisher's suggestions were small, 38 tracked edits that were mostly commas, colons and single words. Across 13 published pages, the comparison from delivery to publication changed about a twelfth of what I'd delivered: a mean of 8%, from 5% to 12%. I call that the publishing pass, although the saved record can't rule out later rounds with me on every article.

Two rows of marks, one per article. My pass changed between 12% and 66% of the words, a mean of 27% across ten articles, plus 7% and 8% in my later pass. The publishing pass changed between 5% and 12%, a mean of 8% across thirteen pages.

SHARE OF WORDS CHANGED, ONE MARK PER ARTICLE
MY PASS
10 articles · 12% to 66% · mean 27%
MEAN 27%
THE PUBLISHING PASS
13 pages · 5% to 12% · mean 8%
MEAN 8%
0%10%20%30%40%50%60%70%
0%25%50%70%
my pass
my later pass, after the publisher's comments (7% and 8%)
the publishing pass
My pass changed about a quarter of the first draft's words, and its spread runs from 12% to 66%. The publishing pass changed about 8% of what I delivered, and every page sits in a narrow band.

My pass rewrites more than it lengthens. Of the sentences in my delivered text, 52% are first-draft sentences word for word, 29% are first-draft sentences I edited, and 19% are new. I dropped 514 first-draft sentences, and the total length barely moved, from 27,831 words to 28,060. Some habits shift the way I'd want them to. Uses of "own" fell by half, from 47 to 23, and uncontracted forms fell from 181 to 143. Others shift the wrong way. Words the floors list as filler rose from 56 to 81, most of them intensifiers like "actually" and "simply" plus a handful of stock transitions, and em dashes went from one to four.

What my pass adds starts with getting the story true. One story in a first draft closed, "That proposal won." It hadn't, so the line became "The proposal didn't win that particular client." The same draft put the story's timeline at "three to four years." The record says three, so the sentence says three. The same habit shows across the set: "year by year" became "month over month", vendor superlatives got hedged and attributed, and an unsourced figure gave way to a sourced one.

Then comes the detail only I hold. A line about the chatbots I built before language models arrived had no count, and I added one: dozens. A chatbot the first draft called "a novelty" became "something of a meme." A comparison between "a wind-up toy" and a self-driving car became a wooden wind-up child's toy against a self-driving car with "heated and cooled massage seats." Those details came from my work and nowhere else. The full-book review of the approved finals found strengths at a higher rate in my words than in the first-draft words I kept: 7.0 to 10.2 per 1,000 words in mine, against 5.2 to 6.9.

Three before-and-after pairs. "I built with tools like ManyChat" gains a count: "I built dozens." "A novelty" becomes "something of a meme." "A wind-up toy" becomes a wooden wind-up child's toy against a car with heated and cooled massage seats. Words I wrote carried 7.0 to 10.2 strengths per 1,000 words, against 5.2 to 6.9 in the first-draft words I kept.

WHAT MY PASS ADDS: THE DETAIL ONLY I HOLD
FIRST DRAFT
I built with tools like ManyChat
MY PASS
I built dozens using tools like ManyChat
FIRST DRAFT
a novelty bolted onto X
MY PASS
something of a meme bolted onto X
FIRST DRAFT
a wind-up toy
MY PASS
a wooden wind-up child's toy
against a self-driving car with
heated and cooled massage seats
STRENGTHS PER 1,000 WORDS, FULL-BOOK REVIEW OF THE APPROVED FINALS
Words I wrote
7.0 to 10.2
First-draft words I kept
5.2 to 6.9
My pass put in details no draft could have: a count, a sharper word and a pair of specific objects. Words I wrote carried strengths at a higher rate than the first-draft words I kept.

My pass also cuts the word that adds force and nothing else, so "depends entirely on" became "depends on." In my later pass a bare "this" got its noun, and "this created new problems" became "the change created new problems." Edits of those kinds come back when the line edit reaches adverbs, pronouns and padding.

The same record shows where the faults sit. The full-book review of my delivered text flagged errors and clarity problems sentence by sentence, and the script traced each one to the earliest version that held its words. Of the faults it could trace, 184 of 215 entered in one of my passes by the strict count, which credits the last version to touch any of the fault's words. By the lenient count, which credits the version where most of those words first appeared, it's 155 of 211. That's 86% and 73%. Counting errors alone, 37 of 38 and 33 of 38. Before anyone relied on the script, it was checked against a hand trace of two articles, and the two agreed on which version held the words for 30 of the 33 faults. As a rate, the full-book review of the approved finals puts 1.6 to 2.9 faults in every 1,000 first-draft words I kept, and 17.0 to 19.0 in every 1,000 words I wrote. Those numbers come with conditions. The first-draft words in them are a filtered set: I'd already rewritten or dropped the weakest lines, so what survived is what I chose to keep. Nobody ever scored the first drafts, so the faults my pass removed appear in no count at all. The numbers locate the faults that remained, and they can't rank one writer against another.

Two bars. By the strict count, 184 of 215 traceable faults entered in my passes, 86%; by the lenient count, 155 of 211, 73%. Words I wrote carry 17.0 to 19.0 faults per 1,000 words, against 1.6 to 2.9 in the first-draft words I kept, which were already filtered by my choice to keep them.

WHERE THE FAULTS THAT COULD BE TRACED ENTERED
STRICT COUNT
credits the last version to touch any of the fault's words
MY PASSES 184
FIRST DRAFT 31
184 OF 215, 86%
LENIENT COUNT
credits the version where most of the fault's words first appeared
MY PASSES 155
FIRST DRAFT 56
155 OF 211, 73%
FAULTS PER 1,000 WORDS
First-draft words I kept
1.62.9
Words I wrote
17.019.0
020
The first-draft words are the ones I chose to keep, and nobody scored the first drafts, so these numbers locate the faults that remained. They can't rank one writer against another.
Between 73% and 86% of the faults that could be traced entered in my own passes, depending on how a fault is credited. The first-draft words in the count are the ones I chose to keep, so it locates faults and can't grade writers.

The faults cluster where I added or lengthened a line. New sentences make up 19% of the sentences in my delivered text and 44% of the sentences where my faults entered, 72 of 162. Rewritten sentences that picked up a fault grew from 16.6 words to 18.7 on average, while across all the sentences I edited the growth was 14.7 to 15.0. Lines I added or stretched held most of my slips.

A fresh patch of dark asphalt sits in an old, worn but unbroken road, and a thin new crack runs along the seam where the new asphalt meets the old.
Faults clustered where I added or lengthened a line. New sentences were 19% of my delivered text and 44% of the sentences where my faults entered.

Faults enter in five recognizable ways, and each has a pair on record, the first draft's line beside mine.

One word that carried the meaning gets swapped. In the first draft, a feature updates its model list "as providers release" new ones. In mine it updates "as providers are released," which makes the companies the thing being released.

A subject changes and its pronoun stays behind. I rewrote a singular subject as a plural one and left the singular "so much as it makes" a few words later, still pointing at a subject that no longer existed.

A frame or a flourish gets added around the point. The first draft said today's tools "already do useful work" and stopped there. I added "countless professionals worldwide" finding new and creative ways to create value, a claim nobody had counted.

An added clause lands on the wrong word or the wrong moment. In the first draft, a proposal template "closed" more than half a million dollars. My version said "having closed," which puts the result before the reuse that produced it.

A plain line turns into explainer phrasing. A clause saying the competitive pressure "runs both ways" turned into the version I wrote: "indicating that the competitive pressure is not one-sided." It's longer, and it sounds less like a person talking.

Five pairs, first draft above my pass. A swapped word makes the providers the thing released. A pronoun stays singular after its subject turns plural. An added frame claims countless professionals. An added clause, "having closed", reorders events. A plain clause about pressure running both ways turns into longer explainer phrasing.

FIVE WAYS A FAULT ENTERS
SWAPPED WORD
FIRST DRAFT
as providers release
MY PASS
as providers are released
Now the companies are what gets released.
STRANDED PRONOUN
FIRST DRAFT
MCP … so much as it shines a light
MY PASS
MCP servers … so much as it makes
The subject turned plural; the pronoun stayed singular.
ADDED FRAME
FIRST DRAFT
today's tools already do useful work.
MY PASS
… countless professionals worldwide …
A claim nobody counted, wrapped around the point.
MISPLACED CLAUSE
FIRST DRAFT
it closed more than half a million dollars
MY PASS
having closed more than half a million dollars
Puts the result before the reuse that produced it.
EXPLAINER PHRASING
FIRST DRAFT
which tells you the competitive pressure runs both ways
MY PASS
indicating that the competitive pressure is not one-sided
Longer, and it sounds less like a person talking.
Each fault my pass let in takes one of five forms, and each has a pair on record. A line editor who knows the five can search the sentences a person touched for them first.

The publishing pass had a different shape. It left 83% of my sentences word for word. It raised mentions of the publisher's brand from 44 to 63, moved summaries and key takeaways to the top, and added keyword headings and call-to-action boxes. The pattern floors barely moved, from 11.3 findings per 1,000 words to 11.1, though the pass did remove all seven runs where a label was followed by a gloss restating it, and two of three em dashes. Of the 252 errors and clarity problems the full-book review had flagged in my delivered text, 27 left the page, about one in ten, most of them fixed. Another 14 were reworded with the fault still in them, and 209 stayed as they were. The last 2 sit hidden in collapsed FAQ answers. The fixes were mechanical and right: a missing space after a period, a reflexive pronoun, a typo.

The publishing pass also added a dozen errors of its own, each one absent from the text I delivered. A technical term my text had right came out wrong. A figure in a summary contradicted the body of the same article. A question no longer matched its answer. A takeaway dropped a qualifier that the summary and the body both kept, and overstated. A file name lost its capitals. A model version disagreed with the same article a few sections down.

One list item shows every hand at work. In the first draft it matched its neighbours: "Trusting the first answer without questioning it." My rewrite turned it into a full sentence, and I wrote: "You should not blindly trust the first answer without questioning it." The rewrite broke the form every other item shared. The publishing pass put the form back: "Blindly trusting the first answer without questioning it." Mine broke it, and the publisher's repaired it.

One list item in three versions. The first draft matches its neighbours, my pass rewrites it as a full sentence that breaks the list's form, and the final puts the form back. Below, six kinds of new errors the publishing pass added, from a wrong technical term to a model version that disagrees with its own article.

ONE LIST ITEM, THREE HANDS
FIRST DRAFT
Trusting the first answer without questioning it.
MATCHES ITS NEIGHBOURS
MY PASS
You should not blindly trust the first answer without questioning it.
BREAKS THE LIST'S FORM
THE FINAL
Blindly trusting the first answer without questioning it.
FORM PUT BACK
NEW ERRORS THE PUBLISHING PASS ADDED, A DOZEN IN ALL, AMONG THEM
  • A technical term written wrong
  • A summary figure that contradicts the body
  • A question that no longer matches its answer
  • A takeaway that drops a qualifier and overstates
  • A file name that lost its capitals
  • A model version that disagrees with its own article
Two hands changed one list item after the first draft: mine broke its form and the publisher's repaired it. The same publishing pass added a dozen errors of its own, which only a read of the live page can catch.

Two independent full-book readings, one of my delivered text and one of the approved finals, put the same four chapters of the sentence-craft book on top: trimming the fat, words that say something else, unclear antecedents and faulty parallels. They read mostly the same words, so the agreement shows the reading is repeatable more than it shows anything new about the text.

Put together, the record says the words I add carry the voice and the slips at once. The setback, the count, the toy and the massage seats arrived in the same pass as the swapped word and the stranded pronoun. The first-draft words that survive are the words I chose to keep, and they carry fewer of the remaining faults in this review. A buyer who thinks the fix for weak drafts is a good human editor is right about half of it. The editor is the hand that puts in the true setback and the lived count. The same hand drops a clause on the wrong noun. Hiring the editor buys the first half, and a line edit after the editor buys the second. So the line edit belongs after the human pass, on the sentences that pass touched, new sentences first. After publishing, the live page gets its own read against the delivered text, because the publishing pass brings slips nobody upstream could have seen.

The file where I keep that client's corrections already holds the rule. A figure the file had banned came back into one article through an edit of mine, after an earlier round had taken it out, and the publisher asked for its source. The rule written after that round says the file binds every sentence, mine included, and a violation left in place without a flag counts as the defect.

A line editor reads every sentence the way the paying reader will

The line edit runs the sentence checks. A publishing house hands a manuscript to a line editor once its structure is settled and before the copyedit, and that editor reads every sentence as the paying reader will, asking one question of each: is it easy and pleasant to read? In my stack the line edit runs after the draft has settled what each part says, and again after any person edits the text, on the sentences that person touched. New sentences go first, because new sentences are where my slips clustered.

A lamp lights a manuscript page on a desk at night, with small pencil marks in its margin and a pencil across it, beside an empty chair pulled back from the desk.
A line editor reads every sentence as the paying reader will, and marks each place that reader would stall, re-read or misread before changing a word.

The standard I chose for this layer is June Casagrande's It Was the Best of Sentences, It Was the Worst of Sentences, a book on sentence craft for working writers. It runs through an introduction, 21 chapters and three appendices, and every unit comes back to one frame, which she states as "The Reader is king." Her guidelines exist to serve that reader, and she bends them when they don't. I turned the book into a working kit for the line edit: a procedure, a reference for each group of chapters, and a set of my approved sentences to calibrate against. The kit restates the book's moves in plain words and never copies its example sentences.

The kit's rules decide what a line editor may touch. Every edit names the place a reader would stall, re-read or misread, and how the change removes it; an edit that satisfies a rule and leaves the reader no better off gets cut. Every change goes into a log as the before, the after and a one-line reason, marked apply, query or keep, so the author can accept or reject each one. The claims stay the author's. When the doer is unknown, the editor asks who it was instead of naming one. It adds no first-person "I tested" to prop up a claim, sends a bad number back as a question instead of swapping in a plausible one, and leaves a warranted claim at full strength without layering hedges over it. Where a better sentence would need a different claim, a missing actor or a detail the author never supplied, the editor writes a question. The voice stays as well: the first read builds a protect list of the draft's chosen fragments, long builds and asides, and no later edit flattens them. Other people's words don't change. Where the book allows something a floor bans, the floor wins. Each pass does one job across the whole text, and a paragraph that needs rebuilding becomes one question to the author instead of a string of edits that adds up to a rewrite.

Three rows of an edit log. An apply edit hands the staring back from the afternoon to the reader. A query cuts three nouns to one action and asks the author whether "entirely" stays. A keep leaves "Grounding sharply reduces hallucinations" unchanged because "sharply" sizes the effect.

THREE ROWS FROM A LINE-EDIT LOG
BEFORE
AFTER
WHY
MARK
BEFORE
your entire afternoon disappears staring
AFTER
you spend the whole afternoon staring
WHY
The afternoon was doing the staring; the staring goes back to the reader.
APPLY
BEFORE
depends entirely upon how people approach using the technology
AFTER
depends on how they approach it
WHY
Three nouns hid one action. Question for the author: keep 'entirely' if nothing else matters.
QUERY
BEFORE
Grounding sharply reduces hallucinations. It does not erase them.
AFTER
(unchanged)
WHY
'Sharply' sizes the effect and sets up the turn.
KEEP
A line edit logs every change it proposes as before, after and reason, marked apply, query or keep. The author rules on each one, and the keeps protect what already works.

Keep rows are required, at least one per section. The full-book reviews used to calibrate the kit spent a quarter of their notes on strengths, 50 of 199. A full-book review of the text I delivered logged 226 strengths in 818 notes, and a full-book review of the approved finals logged 265 in 860. A keep row names a sentence and what it does for the reader, so my later pass and the next editor leave it alone. Two notes from the first review show the form. A sentence that opened "I once worked with a heavy equipment client" and named the exact gear he sold got a keep, because it sets up a real problem before any jargon arrives. The fragments "Evaluating. Refining." got one, because they slow the reader at the practice the article teaches and each has one obvious completion.

The procedure is one read with no marks, then twelve passes in the book's order with one change. The first read writes each paragraph's point in a line and marks every place the reader would stall, re-read, skim or enjoy it; those marks drive the passes, and an edit with no mark behind it needs a strong reason. The overhaul, which the book saves for its second-to-last chapter, runs first, since trimming a sentence whose frame is about to be rebuilt wastes the edits. After the edits, a second straight read checks the clean copy against the protect list, and a word-level diff confirms that every changed span sits inside a logged edit.

A ladder of thirteen rungs: a first read with no marks, then twelve passes, from putting the point in the main clause through length, literal reads, specific words, modifiers, actors, reference and lists, punctuation, fat and the catches, to a second read and a diff check. A legend explains apply, query and keep.

ONE READ, TWELVE PASSES, ONE JOB EACH
  1. 0
    FIRST READ
    INTRO · CH 1 · CH 21
    Read as the reader, with no marks. Note every stall, re-read and pleasure; list what to protect.
  2. 1
    THE POINT
    CH 1 · 2 · 3 · 20
    Put the news in the main clause.
    THE OVERHAUL RUNS FIRST
  3. 2
    LENGTH
    CH 4
    Untangle crammed sentences. A long sentence by choice stays.
  4. 3
    LITERAL READS
    CH 5
    Hear what each phrase literally says.
  5. 4
    SPECIFIC WORDS
    CH 6 · 7 · 19
    Specific nouns; adjectives and adverbs earn their place.
  6. 5
    MODIFIERS
    CH 8 · 9 · 10
    Each modifier beside what it modifies.
  7. 6
    ACTORS AND VERBS
    CH 11 · 12 · 13
    Give actions to their actors. One tense per scene.
  8. 7
    REFERENCE AND LISTS
    CH 14 · 15 · 16
    Every 'the', 'this' and 'it' lands; every list item fits its frame.
  9. 8
    PUNCTUATION AND ATTRIBUTION
    CH 17 · 18 · APP 2
    Every mark serves the reader. Every em dash goes.
  10. 9
    FAT
    CH 19
    Cut the padding; keep qualifications that carry meaning.
  11. 10
    THE CATCHES
    APP 1 · APP 3
    Its and it's, agreement, typos.
  12. 11
    READ AGAIN
    CH 21
    Read the clean copy straight through. Write the keeps, at least one per section.
  13. 12
    VERIFY
    A word-level diff: every changed span sits inside a logged edit.
APPLY
A meaning-safe edit, made directly.
QUERY
A claim, a missing actor, a new detail or a rebuild, sent to the author as a question.
KEEP
A sentence that works, named so the next hand leaves it alone.
The line edit runs one read with no marks, then twelve passes in the book's order, each doing one job across the whole text. The overhaul moves to the front, and every change is logged as apply, query or keep.

The book's first stretch is about where the reader's attention lands. Its introduction asks what the sentence gives the reader. Naming the actor in the subject slot can make an action easier to follow; a passive can also keep attention on the thing the reader is following. The approved finals drew 9 strengths and 2 fixes there. The best example opened on an AI agent drafting a refund policy that didn't exist, a failure with a cost placed before any definition. The worst made a gerund the subject of a sentence about what a product gives you, when the product was doing the giving.

The chapter titled "Who Cares?" asks the main subject and verb to carry what matters to the reader, and it holds 30 strengths against 6 fixes. "That was conversational AI. It was terrible." puts the reader's bad experience first and lets the verdict land in three words.

The chapter on subordinate clauses puts the news in the main clause and asks each "as", "while" and "since" to mean what it says, because "as" can read as "at the same moment" and "while" carries a sense of time. It drew 21 fixes. The worst hid a useful warning behind the main clause "I have to mention", so the grammar made the mentioning the point. The same chapter catches a recommendation written as "if you want X, Y is the best", which makes Y's quality depend on what the reader wants; the fix states Y's quality, then says who it suits.

The chapter on movable parts treats phrases and clauses as blocks: split, merge or reorder them, and nothing is added or lost, which makes these the safest edits a line editor makes. Only 4 fixes landed there. Two clipped lines about a coding tool, "Made up function names" and "Referenced variables I never created," read as lists of things until each got its subject back: it made up function names, and it referenced variables.

The chapter on long and short sentences treats length as a tool, and Casagrande's own line is "Brevity is a tool." Short sentences at the turn are my strongest measured habit, with 46 strengths and 4 fixes. The best example turns its argument on a four-word sentence, "It came from driving," right after a longer one about where that knowledge didn't come from. One apply edit there went the other way: a sentence made the reader hold four question clauses as its subject before the verb arrived, and bringing the verb forward kept every fact. A long sentence by choice stays. One about proposals warns against leaning "too far into vision" in its first half and "too far into spreadsheets" in its second, so its two halves mirror the trade-off they describe, and it goes on the protect list.

Four cards. A gerund subject gives way to the product, "Claude Code provides." A warning hidden behind "I have to mention" becomes the main clause. Four question clauses move behind the verb. A short sentence, "It came from driving," is marked keep.

FOUR MOVES FROM THE BOOK'S FIRST STRETCH
ACTOR INTO THE SUBJECT SLOT
INTRODUCTION
BEFORE
Using Claude Code provides approximately ten times the effective credits
AFTER
Claude Code provides approximately ten times the effective credits
WHY
The product does the providing, so the product takes the subject slot.
NEWS INTO THE MAIN CLAUSE
CH 2
BEFORE
… I have to mention that it is still far from a silver bullet.
AFTER
It's still far from a silver bullet.
WHY
The main clause was the mentioning. Now it's the warning.
VERB BROUGHT FORWARD
CH 4
BEFORE
[four question clauses] … all get clearer once you separate the word from the capability.
AFTER
Separate the word from the capability, and the rest gets clearer: [the four questions].
WHY
The reader held four clauses before the verb arrived. Every fact stays.
SHORT SENTENCE AT THE TURN
CH 4
KEEPThat knowledge didn't come from reading a manual. It came from driving.
WHY
Four words turn the argument, so no later pass touches them.
Four moves from the book's first stretch, each on a sentence from text I delivered: hand the subject slot to the actor, promote the news to the main clause, bring the verb forward, and keep a short sentence where the argument turns.

The chapter on words gone wild asks the editor to hear what each phrase literally says. It holds more errors and clarity problems than any other unit, 15 errors and 31 clarity items among 76 fixes. Both of my examples there came from my pass. A first draft called checkability "most of the battle"; my pass made it "most of the challenge," in an article whose takeaway calls checkability most of the value. Another first draft described chatbots as decision trees "hiding in a trench coat"; my pass made them decision trees "disguised as chatbots", which disguises the chatbots as themselves. Each swap read fluently and emptied or reversed the point.

Two cards of three versions each. "Most of the battle" became "most of the challenge" in my pass, and the fix makes it "most of the value". "Hiding in a trench coat" became "disguised as chatbots", and the fix makes it "decision trees in disguise".

TWO SWAPS THAT CHANGED THE MEANING
FIRST DRAFT
most of the battle
MY PASS
most of the challenge
THE FIX
most of the value

'Challenge' says checkability is the hard part. The article's own takeaway calls it most of the value.

FIRST DRAFT
hiding in a trench coat
MY PASS
disguised as chatbots
THE FIX
decision trees in disguise

'Disguised as chatbots' disguises the chatbots as themselves.

Two words swapped in my pass, and each sentence came to say something the article didn't mean. Each fix matches what the article meant.

The chapter on words gone mild asks for the specific thing, and it's my largest store of strengths: 53, against 33 fixes. The best is a bar regular who switches from bourbon to water around 11 pm, a detail that makes ambient awareness concrete in one clause. One of my edits belongs here: "post-compaction" became "after compaction", a plain phrase where a jargon compound stood. The worst sentence in the chapter asked readers to "ensure you understand its implications", and "its implications" names nothing a reader can act on. The book's demonstration rewrites invent vivid details to show the move. A line editor never does: if neither the draft nor the author supplies the detail, the editor asks for it.

A small wooden wind-up toy car sits on a workbench in front of a modern car's plush seat, lit half warm and half cool.
Specific beats vague. My pass turned a first draft's "wind-up toy" into a wooden wind-up child's toy, set against a self-driving car with heated and cooled massage seats, a detail that came from me and nowhere else.

The chapter on adverbs narrows the folk rule to manner and degree adverbs, the ones that repeat a verb or shout over it. Adverbs drew 23 fixes, nearly all of them polish, and the full-book review of the text I delivered had found "confidently" seven times in one article. I made one there myself: "Agents will absolutely lie" became "Agents will lie", and the bare verb lands harder. "Grounding sharply reduces hallucinations" keeps its adverb, because "sharply" sizes the effect. An adverb that carries a claim, like "safely" or "entirely", goes to the author as a question, since cutting it changes what the sentence promises.

Those edits of mine come from one final pass on one article. Of its 45 changes, 24 were mine, made on a third read after two automated checks and the client's editor had cleared the piece. They're the strongest examples the kit holds, because they show what I fix when nothing prompts me. Three of the four chapters that hold most of the line edit's fixes are still ahead in the book.

Padding, pronouns and broken lists hold most of what the line edit fixes

Both full-book reviews ranked the book's units by fixes and put the same four on top: trimming the fat (106 fixes in the review of the approved finals), words that say something else (76), unclear antecedents (66) and faulty parallels (48). In the traced cases most of those fixes sit in words from my passes: 25 of 31 for words that say something else, 21 of 33 for unclear antecedents, 20 of 27 for faulty parallels and 18 of 23 for fat. Of the four, unclear antecedents is the one where the first drafts contribute a real share, 10 of 33, so the pronoun pass runs on the whole text. The rest of the book follows in the groups an editor runs it in.

Bars for the 25 units of the sentence-craft book by fixes in the full-book review of the approved finals, with the earlier review's count beside each. Trimming the fat leads with 106, then words that say something else 76, unclear antecedents 66 and faulty parallels 48. Breaking the rules draws no fixes and 16 strengths.

ALL 25 UNITS OF THE BOOK, BY FIXES
fixes, full-book review of the approved finals
fixes, full-book review of the text I delivered
strengths, approved finals
Trimming the fat (ch 19)
106
· 4 strengths
Words that say something else (ch 5)
76
· 4 strengths
Unclear antecedents (ch 15)
66
· 1 strengths
Faulty parallels (ch 16)
48
· 41 strengths
Specific over vague (ch 6)
33
· 53 strengths
The major overhaul (ch 20)
28
· 0 strengths
Adverbs (ch 7)
23
· 3 strengths
Tense (ch 12)
22
· 10 strengths
Subordinate clauses (ch 2)
21
· 6 strengths
Nominalizations (ch 13)
21
· 1 strengths
"The" (ch 14)
19
· 1 strengths
Punctuation (appendix 2)
19
· 3 strengths
Semicolons and parentheses (ch 17)
18
· 7 strengths
Prepositional phrases (ch 9)
17
· 0 strengths
Danglers (ch 10)
17
· 3 strengths
Grammar (appendix 1)
16
· 2 strengths
Passives (ch 11)
11
· 10 strengths
Relative clauses (ch 8)
8
· 9 strengths
Deadliest catches (appendix 3)
7
· 0 strengths
Who cares (ch 1)
6
· 30 strengths
Phrases and clauses (ch 3)
4
· 3 strengths
Short versus long (ch 4)
4
· 46 strengths
Attributions (ch 18)
3
· 3 strengths
Introduction
2
· 9 strengths
Breaking the rules (ch 21)
0
· 16 strengths
Two full-book reviews read all 25 units of the book against my articles and put the same four on top. Most of the line edit's work sits in trimming the fat, words that say something else, unclear antecedents and faulty parallels.

Three chapters on modifiers share one fact: a reader attaches a modifier to the nearest word that can take it. In the chapter on relative clauses (8 fixes, 9 strengths), a "that" clause grabbed the nearest noun, so the answers became the thing that "had no idea" about the machines' specs, when the assistant was the one that didn't know. Giving all three verbs one subject fixed it. In the chapter on prepositional phrases (17 fixes and no strengths), a story "I shared at the beginning of this article about" a project seemed to be about the article. In the chapter on danglers (17 fixes), my pass rewrote a plain line into one where "your entire afternoon disappears staring," and the afternoon does the staring. My later pass went the other way: a participle fragment, "Breaking it down file by file", became "I broke it down file by file", and the reader learns who did the work.

A magnet hanging on a string has swung to a small grey nail close beneath it, while a larger amber key it might have lifted lies farther away, untouched.
A reader hangs a modifier on the nearest word that can take it, whatever the writer meant. In "your entire afternoon disappears staring" at a blank screen, the afternoon does the staring.

The next three chapters give actions back to their actors. The chapter on passives (11 fixes, 10 strengths) keeps a passive by choice, when the doer is unknown or beside the point, and rejects one that hides a doer the reader wants. A lead story about three coding agents lost them in "several million tokens of work were done", with nobody doing the work. The editor never supplies a missing doer; it asks the author who did it. The tense chapter (22 fixes, 2 of them errors) wants one tense per scene, and an imagined closed-book test, set in the present, put "took" beside "is answering" in one sentence. The chapter on nominalizations (21 fixes) digs out actions buried in nouns. "The ability to get value out of the AI depends entirely upon" how people use it hides one action in three nouns, and the recast gives it a person and a verb, with "entirely" sent back to me as a question.

Reference is where the line edit finds its biggest class of clarity problems. The chapter on "the" (19 fixes) starts from what the word promises: the reader already knows which thing you mean. An acronym that arrives as "the CTF framework", never spelled out, breaks the promise. The chapter on unclear antecedents (66 fixes, 39 of them clarity) asks that every "this", "it" and "we" point at one noun. "AI doesn't replace reading. This is the biggest misconception," read literally, calls the correction the misconception. My third-read edits gave each bare demonstrative its noun: "All of this happened" became "All of these activities", "this was an older model" became "the agent was an older model", and an opinion line came to open "This assessment is". In a client article the same check runs on "we": the author speaks as "I" and the company as "we", and a "we" can slide from one to the other between paragraphs, most often at a link to another guide. Casagrande's practical repair is to repeat the noun when a pronoun leaves several possible referents.

At a fork in a misty field, a signpost's left arrow points into blank fog and its right arrow points at a lit lantern a few steps away.
A "this" with no noun points into fog. Unclear antecedents drew 66 fixes in the review of the approved finals, 39 of them for clarity, and each fix gives the pronoun its noun.

The chapter on faulty parallels (48 fixes, 5 of them errors) tests a list by distributing its shared frame across every item and reading each result aloud. A list of myths ran "A specific structure. A magic phrase." and then a third item that turned into a full sentence asserting the myth. A sentence promised grounding could "keep sensitive data private" and, two items later, "sometimes never leave your systems", and "keep data never leave" fails as soon as it's read. The repair gives all three items one frame: data stays private, gets processed in a controlled environment, and sometimes never leaves your systems. The same chapter holds 41 strengths, my third-largest store: the slips stand out because the good lists are good.

Two lists tested by distributing their shared frame. In a list of myths, the third item turns into a full sentence asserting the myth, and the repair drops "is". In a promise about data, "keep sensitive data never leave your systems" fails, and the repair gives all three items one frame.

DISTRIBUTE THE FRAME, READ EACH ITEM ALOUD
A LIST OF MYTHS
people who get amazing results … know some kind of secret formula:
people who get amazing results … know some kind of secret formula: … a specific structure
PASSES
people who get amazing results … know some kind of secret formula: … a magic phrase
PASSES
people who get amazing results … know some kind of secret formula: … some obscure scientific knowledge is beyond the reach of common people
FAILS: TURNS INTO A SENTENCE THAT ASSERTS THE MYTH
REPAIR
Some obscure scientific knowledge beyond the reach of common people.
A PROMISE ABOUT DATA
grounding can be built to keep sensitive data
keep sensitive data private
PASSES
keep sensitive data processed in a controlled environment
STRAINS
keep sensitive data never leave your systems
FAILS
REPAIR
so sensitive data stays private, gets processed in a controlled environment, and sometimes never leaves your systems.
Distribute a list's shared frame across every item and read each result aloud. The item that fails is the one that broke the form, and the repair gives every item the same frame.

Punctuation and attribution come next. The chapter on semicolons and parentheses (18 fixes, 7 strengths) asks whom each mark serves. A parenthesis that glosses a term serves the reader; one that crams a GPU count and a permit figure into the middle of a three-verb sentence serves the writer, and the editor gives those facts a sentence of their own. A semicolon that joins two clauses is legitimate and never required, so the editor offers a period as a question and leaves the choice to the author. The chapter on attribution (3 fixes, 3 strengths) defaults to "said", and "admitted" editorializes a company's announcement. The book teaches four uses for the em dash, and the floor bans every one, so each comes out.

Trimming the fat holds the most fixes of any unit. Counted together with the adverbs chapter, 81% of the traced fixes in the two sit in words from my passes. My pass put an announcement in front of a point, "Before we dive deeper", where the first draft had gone straight to the claim. My later pass cut the "own" after a possessive, so "its own answers" became "its answers". Two things look like fat and stay. A qualification that carries meaning stays, and so does a word repeated inside a parallel pair, because the repeat is what makes the pair a contrast.

The overhaul chapter (28 fixes, no strengths) strips a sentence to its subject and verb and reads that kernel alone. A sentence that began "the easiest way to picture the difference" and ended "is like the difference between" says nothing at its core. "A proposal is a modular assembly line" equates a document with the process that makes it, and the rebuild names the process.

The chapter on breaking the rules (no fixes, 16 strengths) says the guidelines bend when the reader is served. Neither full-book review logged a single fix against a chosen fragment, in 818 notes or in 860, and "Evaluating. Refining." stays. The surface still governs: on ads, posts and on-screen text, every sentence needs a subject and a finite verb.

A pruned flowering shrub stands in the center. Two dead grey branches lie cut on the ground beside pruning shears, and three amber blossoms left on the shrub each carry a small blank tag. Far behind, a grey hedge sheared flat has dropped all its flowers.
A line edit cuts the fat and tags what works, at least one keep per section, so the next hand leaves those sentences alone.

The three appendices cover grammar (16 fixes, 6 of them errors), punctuation (19 fixes, 5 errors) and the deadliest catches (7 fixes, every one an error). Grammar caught "Like myself, many people", a reflexive with nothing to point back to. Punctuation caught an example prompt whose opening quotation mark never closed, so the reader couldn't tell where the prompt ended. The catches were a typo in a bold takeaway, a missing space after a period at the top of an FAQ and a fused word in a table cell. All of them sit in the lines a skimmer reads first.

The full-book review of the text I delivered turned all of this into nine habits ranked by payoff per word changed: three points per error, two per clarity item and one per polish item, divided by the words each fix changes. Minutes weren't measured, so the ranking is a proxy for time. The review of the approved finals kept the order:

  • Proofread the lines a skimmer sees: TL;DRs, headings, table cells, FAQ answers and link lines (1.09).

  • Keep one tense per scene (0.60).

  • Read every list item with its lead-in, and reread the whole list when one item changes (0.39).

  • Run the pronoun pass on the whole text (0.37).

  • Read each sentence for what it literally says, starting with every sentence where one word was swapped (0.32).

  • Cut the announcement and the doubled word (0.22).

  • Put each modifier beside what it modifies, and ask who does every added "-ing" word (0.21).

  • Keep parentheses small (0.20).

  • Find the kernel when a plain line turns into an explainer sentence (0.17).

Nine habits ranked by payoff per word changed, from proofreading the lines a skimmer sees at 1.09 down to finding the kernel at 0.17, each with an example, under a band saying the habits run after every hand that touches the text.

RUNS AFTER EVERY HAND

After my pass, after the copywriting pass, and on the live page.

NINE HABITS, BY PAYOFF PER WORD CHANGED
1
Proofread the lines a skimmer sees
1.09
a typo in a bold takeaway
2
Keep one tense per scene
0.60
took beside is answering
3
Read every list item with its lead-in
0.39
the third item in a list of myths
4
Run the pronoun pass on the whole text
0.37
This is the biggest misconception
5
Read each sentence for what it literally says
0.32
most of the challenge
6
Cut the announcement and the doubled word
0.22
Before we dive deeper
7
Put each modifier beside what it modifies
0.21
afternoon disappears staring
8
Keep parentheses small
0.20
a GPU count and a permit figure in one parenthesis
9
Find the kernel
0.17
A proposal is a modular assembly line

Payoff per word: 3 points per error, 2 per clarity item and 1 per polish item, divided by the words each fix changes. Minutes weren't measured, so it's a proxy for time.

Nine habits ranked by how much each fix buys per word it changes. Proofreading the lines a skimmer sees pays most, and every habit runs again after each hand that touches the text.

The kit was checked before I relied on it. A separate model ran it blind on three articles and compared its findings with a full-book review of the same articles. The kit found 84% of what the reviewers found, at about 87% precision, and its misses and over-edits became twelve changes to the kit, among them checks for parallels in tables and list labels, the mechanical punctuation checks, drift in "it" and "we", idioms read literally, and the drone of sibling sections built on one template. Both sides used the same model family, and some of the kit's examples came from those same articles. The 87% precision figure is an estimate from a small sample, with a reported band of 76% to 93%. The comparison measures agreement with those reviewers, not with an audience.

One limit stays. Casagrande's book has no chapter on claims, sourcing or technical precision, and those were 62% of the first review's notes, 848 of 1,368. They stay with the draft's fact check. The line edit makes a piece good to read, and a piece can be good to read and still move nobody to act.

A copywriting pass makes good copy persuade, and it borrows only the structure

The third layer exists to move the reader to act. It runs on the parts of a piece that have to persuade: the headline, the opening, the proof, the offer and the ask. It runs after the line edit, and whatever it rewrites goes back through the line edit, because a rewritten line is a new line, and the trace of my edits showed that new and lengthened lines are where most of the traced faults came in.

If the "conversion copy" you've bought arrived with exclamation marks, a countdown nobody set and an offer the client never made, distrusting this layer is reasonable. The patterns behind persuasive copy are older than any model, and they work on structure: what comes first, what a claim needs beside it, where the ask goes. The sample wording that travels with them carries the exclamation marks and the countdowns, so the pass I run keeps the structure and leaves the wording behind.

Two lists. A copywriting pattern brings its structure across: what comes first, what a claim needs beside it, where the ask goes, and one primary pattern at a time. Its sample wording stays behind: exclamation marks, invented countdowns and offers, the book's unsourced results and its example sentences.

WHAT A COPYWRITING PATTERN BRINGS ACROSS
CROSSES OVER: THE STRUCTURE
  • What comes first
  • What a claim needs beside it
  • Where the ask goes
  • One primary pattern, at most two supporting
STAYS BEHIND: THE SAMPLE WORDING
  • Exclamation marks
  • A countdown nobody set
  • An offer the client never made
  • The book's results, which are its own and unsourced
  • The book's example sentences

The words that fill the structure come from the client's own material.

A copywriting pattern brings its structure across and leaves its sample wording behind. The words that fill the structure come from the client.

The library behind it holds 103 patterns, restated in my own words from Sean Vosler's 7 Figure Marketing Copy and sorted into eleven categories. Each pattern is a short file: the job it does, when to open it, its steps, and the checks that show it was applied rather than named. A pattern supplies the shape of a line. The words come from the client's own material, or from mine on my own pages, and the rule for every headline, ask and card title is to find a real line first and compress it before any pattern touches it. A headline filled from the pattern alone fills from the genre, and the genre is what the buyer was paying to get away from.

Choosing comes before applying. I name the format (a landing section, a post, an email), how much the reader already knows about the problem and the offer, and the one job the piece has. A table gives each job a default pattern and one alternative for a stated condition. I take one primary pattern and at most two supporting ones, and five conflict rules keep them from fighting: one page structure per piece; one curiosity device per screen, paid off on the next; one call to action; loss framing only in the section where the price is; and stakes that land on the situation rather than on the person reading.

The plan for the discovery-call card on my site shows how that reads in practice. The card is sales copy for readers who already know the offer. The job is an offer for a service sold to a business, so the table's alternative applies: the primary pattern shows the buyer's own arithmetic, and the two supporting ones say who should skip the call and what the real catch is. Loss framing got rejected, because no real limit exists for the reader to lose against. The plan ends with one line on what the card has to achieve: an operator with a live problem books the call instead of asking for a proposal, because the card says what the call decides and who shouldn't book it.

Applying ends with checks. Each pattern lists what would show it was only named, with its mechanism missing, and a script runs the block floors over the result. A failed pattern check sends me back to choosing; a floor hit gets fixed where it sits. I look for four shortcuts: a pattern named and not applied, proof or scarcity that nobody recorded, the book's phrasing copied in, and a second primary sneaking in beside the first.

An open tackle box holds eleven lures in eleven compartments, and one amber lure is lifted out above it. Under it, the eleven categories of copywriting patterns are listed with when to open each: mine, hook, reframe, story, prove, desire, frame, offer, structure, channel and show.

ONE PRIMARY PATTERN, AT MOST TWO SUPPORTING
MINE

No real line from the client or the reader exists yet for a slot, or the reader is a guess.

HOOK

Writing or grading a headline, subject line, opener or bullet set.

REFRAME

The reader believes something that makes the offer look wrong or unneeded.

STORY

A fact won't land on its own, or the copy centres the company.

PROVE

A claim needs belief, or the page can't yet turn anyone away.

DESIRE

The copy is accurate, and nobody would want it.

FRAME

Numbers, prices, options or proof counts are on the page.

OFFER

Writing the offer, the price, a proposal or the positioning line.

STRUCTURE

You need the skeleton of a whole page, letter or video sales letter.

CHANNEL

The piece is one of a series, or a channel has its own format.

SHOW

The copy will be laid out on a designed page.

Eleven categories hold 103 patterns, and one rule governs the choice: one primary pattern at a time, picked for the job and the reader, with at most two supporting it.

Each of the eleven categories has a real example in this article's own record.

Three of them supply the raw material and the first line. Mine collects the reader's own words before anyone writes. The buyers further down speak in phrases taken from about 299 saved job posts, each quote string-matched against the post it came from, and an older email engagement for a supplements brand mined customer reviews to write in the language its buyers used. Hook shapes the first line, and its rule is relevance before shock. The dealer's base-setup post, told in full with his rounds below, opens on a question an operator can answer about his own morning: "Does your crew spend 20 minutes every morning setting up a base?" Reframe moves a reader off a belief that makes the offer look unneeded, and it starts from someone who holds that belief. Several job posts quoted below ban generic AI language outright. My own first attempt relied on lists of banned words and phrases. The floors section's replay of 180 flagged lines shows what those lists missed. The pattern states the belief in one positive sentence and what's true in the next, and never opens with the half it rejects.

Two carry belief through a person or a piece of evidence. Story hands over a move through a scene instead of stating it. The owner's call in the opening works that way: he split his own job in two on the phone, and the split is what a reader keeps. Prove earns belief for a claim a reader would otherwise discount, and one of its patterns puts a true admission right before the strong claim. One of the publisher's stories shows it. The first draft closed the story on "That proposal won." I changed the line to "The proposal didn't win that particular client," because it didn't, and let the result follow it: the template behind that proposal went on to close more than half a million dollars of business over the next three years. The setback is what keeps the number from reading like a tidy success story. Proof in this category is found and never made. An admission has to be on the record, and every selling page says out loud who should look elsewhere, as this article's last section does.

A first draft's line, "That proposal won.", becomes "The proposal didn't win that particular client." in my pass, and the result follows it: the template went on to close more than half a million dollars of business over three years.

CONFESSION BEFORE THE CLAIM, APPLIED
FIRST DRAFT
That proposal won.
MY PASS
The proposal didn't win that particular client.
ADMISSION, THEN CLAIM

The template behind it went on to close more than half a million dollars of business over the next three years.

A true admission placed right before a strong claim. The setback is what lets a stranger believe the half-million that follows it.

Two raise the wanting and present the numbers. Desire works on copy that's accurate and that nobody would want, and its core move pushes a feature past its first benefit to the outcome it makes possible. One of the dealer's drafts made its point about the dealer's business model, adding machine control to machines built without it. The rewrite made it about what the operator gets, cutting to grade on the first pass (from the draft section), and every feature in the dealer's later posts had to answer "so what" for the man in the cab. Frame presents facts already on the page in the form that makes a fair decision easy, and it adds none. In the trace above, "about a quarter of the words" and "a mean of 27%" are one fact in two frames: the first is easier to hold, and the second is there for the reader who wants to check it. A frame keeps every material fact in view, and a section carries one framing move at most.

Two shape the offer and the whole piece. Offer names what the buyer gets by its effect and pairs the promise with its real difficulty. The last section does both: what a buyer gets today is the method run by hand on their own work, and a buyer who needs a tool to install with nobody in the loop hears on the first call that this isn't it. Structure gives the order of ideas in a piece, down to the close. The dealer's drafts kept ending on the shop's phone number until one round rewrote 156 of those asks as a single step with a reason from the post's point (told in full with the dealer's rounds).

Before: a dealer's posts ended on the shop's phone number, written for operators who already know the owner. After: "Book a demo and see a receiver pick up corrections with no base on a tripod," one step with a reason from the post.

ONE ASK, BEFORE AND AFTER
BEFORE · 156 REWRITTEN
Ended on the shop's phone number, for operators who already know the owner.
AFTER
Book a demo and see a receiver pick up corrections with no base on a tripod.
ONE STEP, ONE REASON FROM THE POST
One round on an equipment dealer's posts rewrote 156 asks. Each close became one step with a reason taken from the post's own point.

The last two work at the scale of the series and the screen. Channel gives every piece in a series one job and one ask, and lets the platform set the format. A Facebook post I sent back as "written like a tweet" failed this category in a single line, and the invented restock text and win-back email in the buyers section show one voice doing two jobs on two channels. Show makes the image and the marks on a page argue the same claim as the sentence. The picture of a leak upstairs and a bucket downstairs in the draft section acts out its rule, and amber marks the thing that works in every figure on this page, so a reader learns the code once and reads it everywhere after.

The book these patterns come from teaches selling, and parts of it can't cross onto a client's page. Its results and statistics are its own and unsourced, so none of them enters copy as fact; a page uses the reader's sourced numbers or none. Its sample emails invent seat caps and same-day bonuses, and here urgency is a fact about supply or it's absent. Many of its contrasts deny one thing in order to assert another, and each of those gets rewritten as two plain positive sentences. Where the book casts the reader as the hero, it often narrates what the reader feels. This layer describes the situation instead and lets the reader recognize it. The floors section runs both books through the same minimums.

The library hasn't earned a claim of lift. Its test is written: three real pieces, each compared blind against its line-edited version, by a reader who doesn't know which is which. That comparison hasn't run yet. So this section claims a structure and a set of rules, and the first measured result goes into the record whichever way it lands.

A wooden concrete form beside a creek is being filled with smooth river stones gathered from the creek bed, a small stack of them waiting at the water's edge.
A pattern supplies the form, and the client's own words, gathered close to home, fill it. A headline filled from the pattern alone fills from the genre.

One question is left that the three layers can't answer between them. A draft, a line edit and a copywriting pass, run by several hands, still need something that keeps all of it sounding like one client.

A voice compiler makes the same content speak for one client

That something is the voice compiler. It receives a content map where every part already has a job and every claim already has its evidence, and it makes those parts sound like one business talking to one audience on one channel. Before it renders a word, it needs five inputs written down, because a voice nobody wrote down gets replaced by the model's default.

A violet sailboat with a lit lantern follows a curving line of five lit buoys toward its own small dock on the left. Off to the right, a crowd of identical grey boats has drifted into one cove, their lanterns dark.
Who's speaking, the client's profile, the audience, the channel and a reference set get written down before drafting starts. Without those marks, every client's copy drifts to the same place: the model's default.
Input What it holds
Who's speaking Me in my own name, me under a client's byline, or the brand as "we". Each changes the pronouns, who can be quoted, and where a source may appear.
The client's voice profile The words this business uses and avoids, the proof it trusts, what it would never say, how often it contracts, and the usual shape of its sentences.
The audience Who reads it, what they know, and their words for their own problems, collected where they talk when nobody's selling to them.
The channel Length, how a post opens, who owns the page, and what a reader is doing when it reaches them.
A reference set Pieces the client approved for this channel, used as targets and as the test the draft is scored against.

For voice itself, one email-marketing company's public style guide draws the line I use: "Our voice doesn't change much from day to day, but our tone changes all the time." Voice is who's talking. Tone is how that person adjusts to the moment, quieter with a customer whose order went missing than with one who just ordered again.

Nielsen Norman Group breaks tone into four dimensions: formal or casual, serious or funny, respectful or irreverent, matter-of-fact or enthusiastic. Those dials describe a difference between two pieces of copy. They're too coarse to produce one: two equipment dealers can sit at the same setting on all four and sound nothing alike, because the dials say nothing about the words a dealer uses, the proof it trusts, the jokes it would never make, or which customers it's talking to.

The same shopkeeper, in the same amber apron and cap, appears twice at one counter. On the left, in dim light, the shopkeeper leans in quietly beside an empty opened parcel box. On the right, in warm light, the shopkeeper hands over a wrapped parcel with arms raised.
Voice is who's talking, and it stays put. Tone is how that person adjusts to the moment, quieter for a customer whose order went missing and brighter for one who just ordered again.

The sociolinguist Allan Bell studied that last item, and his finding opens his paper on audience design: "Style is essentially speakers' response to their audience." Speakers adjust mostly toward whoever they're addressing, and in mass media, Bell found, they also shift toward an absent group they identify with, a group he called referees. Referees explain why a dealer's posts should sound like the operators who read them in a cab, even though the dealer pays for the writing. Bell's finding also sets the unit a voice compiler works on: one client, talking to one audience, on one channel. Change any of the three and the voice has a different job.

The channel input carries real numbers. The dealer's captions I'd reviewed and kept had a median sentence of 12 words on LinkedIn, 10 on Facebook and 8 on Instagram, measured on 13, 19 and 15 captions. On each channel, at least half the captions put a contraction in every place one could go. Same company, same products, three sentence shapes.

The numbers come with a caveat. They were fit on summer captions I had approved plus eight of my rewrites, 56 in all, and the owner later sent some of those summer posts back as hard to follow. So the bands describe my approved summer posts and can't say whether that shape was right. A measured profile tells you how a voice behaves. It can't tell you whether the audience trusts it, which is why the voice-fit score in the system I'm building advises and never blocks.

Three channel panels for one equipment dealer. The median sentence runs 12 words on LinkedIn, 10 on Facebook and 8 on Instagram, measured on 13, 19 and 15 captions. On all three, the median caption used a contraction everywhere one could go.

ONE EQUIPMENT DEALER, THREE CHANNELS
LINKEDIN12
words in the median sentence
Contractions: used everywhere one could go
13 captions measured
FACEBOOK10
words in the median sentence
Contractions: used everywhere one could go
19 captions measured
INSTAGRAM8
words in the median sentence
Contractions: used everywhere one could go
15 captions measured

Fit on approved summer captions plus eight rewrites, 56 in all. The numbers describe how those posts behaved, not whether readers trusted them.

Same company, same products, three sentence shapes. An equipment dealer's approved captions ran 12 words a sentence on LinkedIn, 10 on Facebook and 8 on Instagram, and contracted wherever they could.

Some features of a voice are stable enough to count. When Frederick Mosteller and David Wallace worked out who wrote the disputed Federalist papers, they did it with the dullest words in the language. Their abstract notes that "filler words of the language such as an, of, and upon" provide "fairly stable rates" for a given writer, whatever the topic. Hamilton and Madison could argue about anything, and their rates for "upon" still gave them away. Contraction rate, function words and sentence-length patterns are features I can compare in a client's writing. Their usefulness has to be checked against that client's material. Resemblance to an author still differs from whether a reader hears the post as one of their own: Mosteller and Wallace could tell Hamilton from Madison, and their method couldn't say whether a shopkeeper in New York found either man convincing.

One studio's copy shows how measurement can help locate a voice problem. My first round of AI-assisted copy for that studio sounded like AI, and the client couldn't say what was wrong with it. I measured the client's own writing across hundreds of assets in nine families of metrics and rebuilt the copy against that profile. The case record reports that the rebuilt campaigns were accepted and that the client could compare the versions in an interactive deliverable. That gives the measurement a useful role without making numerical resemblance a substitute for the client's read.

Who's speaking is the input that went wrong for the dealer on launch day. My standard requires every claim to name who saw it. A writer following that rule literally inside the dealer's own posts attached the owner's name and the interview date to sentence after sentence, in 135 of the 252 posts queued for launch. The fix moved the source behind the post. Before: "[The owner] described the offer in one line in August: 'just show up, drop it on your job.'" After: "We show up and drop it on your job." Same fact, same source, still on file for the team, and the company is now the one talking. My test is whether a trade magazine could print the paragraph about the company without changing a word. If it could, the company is a bystander on its own page.

Before: a post reads "[The owner] described the offer in one line in August: just show up, drop it on your job." After: the post reads "We show up and drop it on your job," and the owner's name and the date sit in a source note below it.

ON THE CLIENT'S OWN PAGE, THE COMPANY SPEAKS
BEFORE
[The owner] described the offer in one line in August: 'just show up, drop it on your job.'
AFTER
We show up and drop it on your job.
SOURCE NOTE, FOR THE TEAM
Owner interview, August: 'just show up, drop it on your job.'
Kept on file. It doesn't run in the post.
On a client's own channel, the company is the speaker. The fact stays the same and the source stays on file for the team, moved out of the post and into a note behind it.

In this article the rule runs the other way: I'm the byline, so a source named in the prose is right, and an unattributed "we" would hide my evidence. The same rule puts the source in opposite places on two channels, so the byline is an input.

Building a profile for a new client starts small. I read at least five pieces the client's audience already engages with, then write a short voice contract: who's speaking, to whom, about what, two words for the tone, one pattern this client must never use and one it must always use. Every approved and rejected draft adds to it, and the measured numbers come once there's enough approved work.

The best evidence I've found for a profile is the record of what a person keeps, adds and cuts. For my own byline, the trace above reads as a profile: I fix what a story claims before how it reads, add the detail only I hold, contract, and cut the word that adds force and nothing else. The same record lists the slips my additions bring, a "this" with no noun or a clause on the wrong noun, and the line edit catches those.

Every layer reads the profile. The draft uses the client's way of speaking and the proof the client trusts. The line edit builds its protect list from it (the chosen fragments, long builds and asides that are the voice) and leaves them standing. The copywriting pass grounds every headline, ask and card title in the client's own lines. The checker counts what it keeps as well as what it flags, so a pass that flattened the client's habits shows up as a falling keep count.

With the inputs set, the work runs in order across the layers. It declares the target first: speaker, audience, channel, and how formal, funny, personal and pointed the piece should be. It drafts each part of the map for that channel, which is where a three-sentence LinkedIn paragraph becomes a two-line Instagram caption carrying the same point. The line edit runs against the profile's protect list, the copywriting pass works on the parts that persuade, and the line edit runs again on what that pass touched. The shared minimums run next, and a voice-fit score against the reference set gets reported without blocking. Then the client, or someone from the audience, reads it as the audience would, and every change gets filed with its reason. When the same correction shows up twice, it becomes a rule in the client's file, and the next draft has to pass it.

Five narrow storefronts stand in a row, built to one identical plan with the same door, window and roofline. Each is dressed differently: an amber awning with ferns, a violet front with a brass lamp, ivy and a bench, a periwinkle door, and a tiled step with lanterns.
Five shops share one plan, and each is dressed its own way. Two clients can share most of a content map, and if their copy reads alike, the voice compiler did nothing.

Two buyers on one channel make the test concrete. Put the dealer's LinkedIn posts next to the LinkedIn posts a founder hires a ghostwriter to write. Every one of them gets the same four-line brief: who it's for, what they get, why they should believe it, what to do next. The voice compiler is where they part. The dealer's posts speak as the company, "we," to an operator reading on his phone in the cab. The founder's posts go out under his own name, as "I," in the voice his audience already knows from him. Their posts can carry the same parts in the same order. They shouldn't share a single opening line. If the two read alike, the voice compiler did nothing.

The first review's advice on modelling my voice applies past my own byline: "The most faithful model of your voice will be built from those distinctions, not from a list of forbidden words." The distinctions come from edits a person accepts or rejects, with the reason attached. A list of banned words can't hold that, and a record of decisions can. Each profile also records how the shared writing checks apply to that client's accepted work.

Every client gets the same minimums underneath

Under every client's profile sit the shared checks I call floors. They set my drafting defaults. A client's accepted work can justify a different choice on that client's pages, and the exception belongs in the client's record. My standard's dial-and-wall comparison describes the relationship: voice varies within written constraints, with any accepted departure made explicit.

A few of the floors, to make that concrete:

  • No em dashes, anywhere.

  • No "it's not X, it's Y" framing, which argues with a belief the reader never held.

  • No "the" in front of something the reader hasn't met yet, such as a product nobody has introduced.

  • Every claim names who saw it: me, a named source, a study, a practice the reader can look up.

  • Every headline is one complete thought a stranger could repeat after one read.

Three voice blocks sit on one shared base. A construction dealer speaks as the company, an AI-tools publisher keeps a deliberate fragment, and a studio contracts most of its verbs. The base lists five floors every client shares.

THE DIAL, SET PER CLIENT
CONSTRUCTION DEALER
Speaks as the company. The owner's name is on file so the checker catches a borrowed byline.
We show up and drop it on your job.
AI-TOOLS PUBLISHER
Keeps a deliberate fragment, like "Evaluating. Refining."
STUDIO
Contracts between 59% and 94% of its verbs. My first round managed 4%.
THE WALL, THE SAME FOR EVERY CLIENT
  • No em dashes
  • No 'it's not X, it's Y' framing
  • No 'the' before something the reader hasn't met
  • Every claim names who saw it
  • Every headline is one complete thought
The voice is the dial and the floors are the wall. A dealer, a publisher and a studio set the dial three ways, and all three stand on the same five minimums, which no client's voice gets to move.

Writing the floors down was the easy part. Getting a model to obey them, even a model that had just written them, is where the process fell apart. The first full version of my writing standard ran about 21,000 words, drafted by a model that had spent a day defining what bad machine writing is. Its opening paragraph broke its own rule against the "it's not X, it's Y" shape, and I caught that on first read, along with a fragment tacked onto the end of the same paragraph for rhythm.

So a separate pass ran the document's own rules over the document, one rule at a time, and logged every change. It made 72 edits to a file that defined every pattern it was fixing. Thirty of them were that same negative contrast. Six were sentences announcing their own significance and five were filler, and three each were empty figures of speech, lines kept only for their rhythm, objects given agency, purple phrasing and internal vocabulary. The writer's check had searched for two surface forms of the negative contrast and walked past the version that spans two sentences.

A person pushes a wide broom down a long corridor toward a glowing doorway. The floor ahead is dusty. The floor behind is swept clean except for a line of muddy bootprints left by the sweeper's own steps.
A model that spent a day defining bad machine writing drafted a standard against it, and a separate pass still made 72 edits to that standard. Thirty of them fixed the negative contrast the document banned.

That was the first lesson the floors taught, and every check I've run since follows it. Knowing what bad writing is doesn't stop a model producing it. The method that removes it runs one rule per pass, by a reader who didn't write the draft, with a log of every change, because a pass that changes one thing can be checked and a rewrite that changes everything can't.

The second lesson came from lists of banned words and phrases, the tool I'd reached for first. An audit of my case studies had produced 270 flagged lines, and 180 of them quoted the offending text cleanly enough to replay. Replayed through the detectors I already had, those 180 lines set off the pattern lists 15 times, and 13 of the 15 were a sourcing check that happened to trip, which wasn't the problem the auditor had named. The strongest detection engine I had fired on none of them.

A grid of 180 squares, one per flagged line replayed through existing detectors. Fifteen squares are lit for pattern-list hits, thirteen of them from an unrelated sourcing check. The strongest engine lit none. Every other square stays dark.

180 FLAGGED LINES, REPLAYED THROUGH MY DETECTORS
180flagged lines replayed
15fired a pattern list, 13 of them an unrelated sourcing check
0fired the strongest engine

Empty squares: no detector fired.

An audit of my case studies flagged 270 lines, and 180 quoted the text cleanly enough to replay through my detectors. Pattern lists fired on 15, mostly for an unrelated reason, and the strongest engine fired on none.

The largest class of real problems in that audit, about 89 of the 270 lines, was a sentence pointing at itself: a line announcing that an insight is coming, a card naming its own category, a caption describing what the figure is instead of what it shows. No word list catches those, because the words are ordinary. The defect is in what the sentence does, so a vendor whose quality control is a banned-word list has automated the part of the job that catches the least.

So the rule list behind my checks has two engines. Patterns decide what a pattern can decide: an em dash, a banned transition, a missing contraction in a voice that contracts. A judge decides the rest, one narrow question at a time. Does this "we" presume a closeness the reader hasn't granted? Does this sentence say anything the paragraph didn't already say? Does this summary keep the body's condition? That judge is a model, which might sound like it contradicts the start of this article, where two model reviewers scored drafts 77 and 88 that I scored around 30. The difference is the question. A broad quality score didn't match my judgment in that trial. A model asked whether one sentence keeps one condition has a narrow job that can be checked. A judged rule only gets to block a draft after it reproduces known answers on examples it has never seen. The statistical signs of machine prose, like even sentence lengths and the same few connectives on rotation, get measured and reported, and no model is asked to judge them. The judge also has to know what to leave alone. Keep signals sit in the same list: a concrete scene, a candid setback, a fragment whose missing piece is obvious, an analogy that shows a mechanism. Without them, a checker trained to find defects finds them everywhere and sands the client's voice down to the model's average.

Two engines. Patterns decide 21 rules, such as an em dash or a banned transition, and block on a hit. A judge answers 32 narrow questions, such as whether a summary keeps the body's condition, and warns until each rule is cleared. Keep signals, such as a concrete scene or a candid setback, sit in the same list.

TWO ENGINES READ ONE RULE LIST
PATTERNS

decide what a pattern can decide

  • An em dash
  • A banned transition
  • A missing contraction in a voice that contracts
21 RULES

block a change on a hit

A JUDGE

one narrow question at a time

  • Does this 'we' presume a closeness the reader hasn't granted?
  • Does this sentence say anything the paragraph didn't already say?
  • Does this summary keep the body's condition?
32 RULES

report and warn until each is cleared

KEEP SIGNALS, IN THE SAME LIST
  • A concrete scene
  • A candid setback
  • A fragment with one obvious completion
  • An analogy that shows a mechanism

A judged rule gets to block only after it reproduces known answers on examples it hasn't seen. Two more rules read the page as it went live.

Patterns decide what a pattern can decide, and a judge answers one narrow question at a time. Keep signals sit in the same list, so the checker knows what to leave alone.

The last lesson was about where rules live. When I went looking, the banned-word list existed in at least five copies across my projects, in code and in prose, and the copies disagreed. A correction made in one copy had no way to reach the others, so I consolidated them into one list that every check reads. The first review fed it directly: its findings on claims became six checks with plain names. One catches a capability stated as a guarantee, one a claim that slides from one thing to another (a model to the app built on it), one a summary that drops the body's condition, one an anecdote stretched into a claim about everybody, one a contrast drawn wider than the real distinction, and one a promotion that arrives before any proof. A prose document can explain a rule; only the list defines it.

Both books the layers borrow from pass through the same wall, and where either allows something a floor bans, the floor wins. The sentence-craft book suggests fronting a clause that the floors want behind the main clause in conversion copy, a teasing "the" at the opening of a story, and a passive that asserts a reputation without saying who holds it. Its advice to cut qualifications when in doubt stops at any qualification that changes the claim, and some of its own transitions sit on the kill list. Each of those conflicts is written down with its resolution, so the line edit never settles one case by case. The copywriting book gets the same treatment: its unsourced results, its invented scarcity and its habit of denying one thing to assert another never reach a page.

Two books lean against a violet stone wall that holds them up. Under the picture, six rows show what the sentence-craft book and the copywriting book allow and what the floor does instead: em dashes replaced, fragments given verbs, a first "the" introduced, invented details sent back, invented scarcity cut, and denial contrasts rewritten as two positive sentences.

THE BOOK
WHAT IT ALLOWS
WHAT THE FLOOR DOES
SENTENCE CRAFT
WHAT IT ALLOWS

Four uses for the em dash

WHAT THE FLOOR DOES

Every em dash becomes a period, comma, colon or parentheses.

SENTENCE CRAFT
WHAT IT ALLOWS

Fragments on ads and on-screen text

WHAT THE FLOOR DOES

Every sentence there gets a subject and a finite verb.

SENTENCE CRAFT
WHAT IT ALLOWS

A teasing "the" at an opening

WHAT THE FLOOR DOES

A first mention gets "a", a possessive or a description.

SENTENCE CRAFT
WHAT IT ALLOWS

Invented vivid details

WHAT THE FLOOR DOES

Only details the author supplied; anything else goes back as a question.

COPYWRITING
WHAT IT ALLOWS

Invented scarcity: seat caps, same-day bonuses

WHAT THE FLOOR DOES

Urgency only as a real fact about supply.

COPYWRITING
WHAT IT ALLOWS

A contrast that denies one thing to assert another

WHAT THE FLOOR DOES

Two plain positive sentences.

Where either book allows something a floor bans, the floor wins. The sentence-craft book's em dashes and the copywriting book's invented scarcity both stop at the same wall.

A per-client block is where the floors meet the voice compiler. Each client gets a short written list of exceptions built from their own approved pieces. The dealer's block notes that its pages speak as the company, with the owner's name on file so the checker can catch a borrowed byline. The publisher's block licenses a deliberate fragment its accepted work uses, like "Evaluating. Refining." The studio's block records its contraction rate. The checker reads a client's block before it flags anything, so the publisher's licensed fragment passes on the publisher's pages and fails on the dealer's, whose post rules require a subject and a verb in every sentence. The wall is the same for everyone, and the exceptions are written down per client.

On a client's property, the client's accepted work outranks a house floor where the two disagree. The floor still flags the line, and the client's record decides. The dealer's posts put that rule to its strictest test.

Construction posts went through more than a dozen rounds

The owner's rule is that no post at all beats one his customers can tell was written by someone who doesn't understand their work. The readers are machine operators and contractors, many of them between 45 and 65 by the owner's account, often reading on a phone in the cab. When he went looking for a writer, his job post asked for a professional, industrial tone with no marketing fluff. My first note on the old copy asked for the same thing in my words: "We have to be able to speak direct like an actual construction operator out on a job site, or if some gentlemen ran into each other at a bar, how would they speak?" The platform side of that engagement, where an agent platform drafts posts from vendor brochures and spec sheets, recorded interviews and support stories, is written up as a case study. The social rounds are the part it doesn't cover, and each round below fell to one layer of the stack.

The owner's first feedback came on a batch of summer posts. He approved five and sent twelve back with notes. Six of the twelve said he didn't understand a sentence, and five of those six landed on the third sentence or the third paragraph. My reading of that pattern is that sentence two wasn't carrying the reader into sentence three, so by the third line he was guessing. The fix belonged to the line edit, the layer that reads how each sentence hands off to the next: every sentence has to carry the one before it forward, because confusion that starts small has compounded by line three.

A timeline of rounds on a dealer's social posts, each tagged with the layer that owned it: summer send-backs for the line edit, a launch-day byline sweep of 135 of 252 posts for the voice compiler, 156 calls to action for the copywriting pass, new post rules, majors cut from 10 of 33 to none, write passes shrinking from 33 captions to 1, and a queue waiting on the owner's ear.

MORE THAN A DOZEN ROUNDS, ONE CHANGE EACH
  1. SUMMER SEND-BACKS
    5 approved, 12 sent back. 6 of the 12 said a sentence didn't make sense.
    LINE EDIT
  2. LAUNCH DAY
    A borrowed byline in 135 of 252 queued posts.
    VOICE COMPILER
  3. CALLS TO ACTION
    156 asks rewritten.
    COPYWRITING PASS
  4. POST RULES
    A subject and a verb in every sentence, the four-line brief, nine named slop patterns.
    FLOORS + DRAFT
  5. REVIEWER ROUND
    10 of 33 captions rated major. None after one revision.
    FLOORS
    100
  6. WRITE PASSES
    Fewer posts needed rewriting each pass.
    FLOORS
    33
    14
    7
    2
    1
    CAPTIONS PER PASS
  7. A LATER BATCH
    Majors fell from 8 to 3 after its rewrite.
    FLOORS
    83
  8. THE QUEUE
    Calendar pushed back twice. The posts wait on the owner's read.
    THE EAR
Each round on an equipment dealer's posts changed one thing and kept the change as a rule. Major problems fell from 10 of 33 captions to none, and each write pass touched fewer posts than the last.

Launch day was the voice compiler's round. After 135 of 252 queued posts turned out to quote the owner like a news source, or cite the interview, on his own page, the borrowed-byline rule from the voice section went into his block.

The week after launch was the densest, and it opened with the copywriting pass on the calls to action. The drafts kept ending with the shop's phone number, and my note on that was blunt: putting the number in "is insulting," because most of his customers already have the owner's cell number. The asks got rewritten to one step an operator would act on, like swinging by the shop for a demo, with a reason taken from the post's own point. That round rewrote 156 of them.

Then came the post rules, because clean sentences were still producing posts I called slop, and each rule landed in a different layer. The first is a floor for posts: every sentence has a subject and a finite verb. An opener of bare fragments, "Every morning" and "Every crew" stacked after a number, fails, because a cold reader can't tell what the twenty minutes is doing. The second belongs to the draft: the four-line brief from the content section, with the point stated first, as a benefit the viewer gets.

The third rule became a list, split between floors a script can check and calls a judge has to make. Every time I said "that's slop" about a post with clean grammar, the pattern behind the verdict got found, named and written down, so every later post could be checked against it. Nine patterns came out of that week. A line like "Five crews lose 166+ paid hours a season" presumes the reader runs five crews; he might run one. A post opening "One tiltrotator can work on more than one of your excavators" answers a question nobody asked yet. A line like "When crews decide, they want it yesterday" narrates the seller's private knowledge about how buyers buy, to the buyers.

Nine cards, one per named pattern, each with an example from a real draft: presumed reader, answer with no question, purpose stated last, transcript stitching, caveat that talks down the product, disconnected assertions, no stated benefit, throat-clearing lead-in and leaked sales playbook.

NINE NAMED PATTERNS FROM ONE WEEK OF POSTS
01Presumed reader
"Five crews lose 166+ paid hours a season."
PASSES: Frame it as a hypothetical: "Say you run five crews."
02Answer with no question
"One tiltrotator can work on more than one of your excavators."
PASSES: Open on the buyer's real question.
03Purpose stated last
"click one button, pull back on the stick, machine control cuts grade"
PASSES: Say what it's for, then how.
04Transcript stitching
"is for light material placement like that"
PASSES: Keep the client's facts and write new sentences.
05Caveat that talks down the product
"a huge highway job still needs a motor grader"
PASSES: Say what the product is for.
06Disconnected assertions
A list of attachments with no reason to care.
PASSES: Tie the list to the reader's use.
07No stated benefit
"I don't understand the benefit of this."
PASSES: State the benefit before the ask.
08Throat-clearing lead-in
"With machine control, the machine…"
PASSES: Put the subject and verb first.
09Leaked sales playbook
"When crews decide, they want it yesterday."
PASSES: Say what the product does and what the viewer gets.
Every post I called slop had clean grammar, so each verdict got traced to a pattern, named and written down. Nine came out of one week, and every later post was checked against all nine.

The remaining drafts then went through the rounds behind the numbers in the opening. A reviewer model that hadn't written any of them graded the first full rewrite of 33 captions and rated 10 as major problems. After one revision, it rated none that way. Four of those ten had brought back wording I'd already cut in an earlier round, the repeat-correction problem in its plainest form, and fourteen of the 33 calls to action ended on the same word, "yourself."

The reviewer checked facts as well as wording, which is the draft's job of matching every claim to its evidence. On a later batch it caught a post that described a trench job as one person's work from the cab. The owner's recorded words said otherwise: one of the two people in the trench is freed up, and the other stays to push the joint together. It also caught three near-copies of posts already scheduled, down to whole sentences.

Each of my edits became a named rule, and every remaining post had to pass it. Here are a few, in the order they showed up, and between them they touch every layer.

One end card told operators to put their own hands on the sticks. Read by a heckler, it says something else, and I couldn't publish it. The fix invited them to book a demo and run a cut themselves. The heckler test is the line edit's literal read applied to a whole post, and a sweep of 297 captions found that line was the only instance.

The remote-support opener from the content section came next, the one that cast the reader as a helpless victim waiting for someone to drive out. Its rule went into his voice profile: open on the situation, and let the trouble belong to the machine, not to the person reading.

A long workbench has four wooden stops clamped along its fence, the newest one amber, and a plank slides past them. At the left end lies a grey pile of crooked miscut boards. At the right end sits a neat stack of square-cut boards.
Each edit I made to an equipment dealer's posts became a named rule that every remaining post had to pass. One sweep checked 297 captions against a single new rule and found one line to fix.

A post corrected a belief the reader might hold, and it came out confrontational. I compared it to "trying to convince someone to read the Bible by beating them with it." The fix sits with the copywriting pass, which puts the reader's belief and what's true in two plain sentences. The rule: offer relief, and never tell the reader what he believes.

A post listed a feature and stopped. My note: "The 'so what' has to be the central focus of ANY post." Another said the reader would have someone to call, and the copywriting pass now asks every post one more question: what, specifically?

Some of my edits were single words, and those went to the voice profile. "Model" became "specifications," because that's the buyer's word for it. A support line became "supports your metal remotely," because "metal" is what operators call their machines.

Others were about the whole post. A draft built around a statistic on operator job openings, most of them replacements, stated the number and stopped there. My note was the draft's sequence in one line: find the point you're trying to make, then "start from there, then write it, then ask yourself if what you wrote meets the point." A Facebook post came back from me marked "written like a tweet," the channel profile from the voice section failing in a single post. And one post lost its meaning in a shortening pass, because the cut removed the one step that connected the problem to the product.

One post shows most of this at once. Its first draft opened, "Every crew spends 20 minutes setting up a base every morning." That sentence states a fact about every reader's crew. The reader might not run a base station at all, and the line reads as the dealer telling him about his own morning. The rewrite asks: "Does your crew spend 20 minutes every morning setting up a base?" Then it states what the product does: it skips that setup by sending corrections over a cell connection, so there's no base on a tripod. Then the ask.

In August, the owner had approved an earlier version of the same idea that opened "20 minutes setting up your base. Every morning. Every crew." His approved post breaks two of my floors, the presumed reader and the verbless sentence. On his page, his accepted work outranks a house floor, so the floor flags the line and his record decides. The rewrite went into the queue anyway, because it carries the idea and the numbers he approved without the presumption. His approval is also why his ear stays in the loop: the floors can be wrong about what his audience accepts, and only his read or theirs settles it.

A four-line brief. Audience: crews that set up a base station every morning. Point: you can skip that setup. Proof: corrections arrive over a cell connection, so there's no base on a tripod. Ask: book a demo. Below it sits the post's opening question.

THE FOUR-LINE BRIEF, FILLED IN
AUDIENCECrews that set up a base station every morning.
POINTYou can skip that setup.
PROOFCorrections arrive over a cell connection, so there's no base on a tripod.
ASKBook a demo and see a receiver pick up corrections with no base on a tripod.
THE OPENING LINE IT PRODUCED

"Does your crew spend 20 minutes every morning setting up a base?"

Before a dealer's post about base setup got rewritten, it got a four-line brief: who it's for, what they get, why they should believe it, and what to do next. The opening question came out of that brief.

That's where the rounds left the queue: through every check a rule or a cold reader can run. The calendar was pushed back twice so the owner would have time to read it properly. My account of the latest round was that he no longer saw AI slop or grammar problems. The saved call records his plan to read for sense and voice, rather than blanket approval of that batch. The countable work spans more than a dozen rounds; an exact total depends on which writes and reviews count as a round. What's left is the read he said he'd make from the start, whether the posts make sense to an operator and sound right, and no floor measures that.

An empty excavator cab at dusk, a phone glowing in its dashboard mount, the bucket resting outside the windshield beside a freshly graded pad.
An operator reads an equipment dealer's posts on his phone in the cab. The owner reads each one for that operator, and his ear settles what no floor measures: whether it makes sense and sounds right.

The owner reads for that operator, and his ear is the last step of the stack.

Six kinds of buyers hit the same split in different places

Partway through my work for the AI-tools publisher, I put one line at the top of the file where I keep that client's corrections: "If I have to go and make the same correction every single time, then I might as well just write it myself." That line is the economics of every engagement below. The file of corrections, each with its reason, lets the next draft be checked for the same problem, because every correction becomes a rule the next draft has to pass.

The situations below come from job posts I've read, and where a buyer's own words say it best, I've quoted the post. Each buyer leans on a different layer.

An agency writes for a whole niche of clients

A shared page map with five parts sits across the top: who it's for, the problem in the customer's words, what the contractor does, the proof and the ask. Below it, an HVAC, a plumbing and a roofing client each keep a voice profile and a file. One owner's edit goes into that client's file only.

SHARED PAGE MAP, BUILT ONCE FOR THE NICHE
  1. Who it's for
  2. The problem in the customer's words
  3. What the contractor does about it
  4. The proof
  5. The ask
HVAC CLIENT
Voice profile from its reviews, its front desk, the owner's recorded talk and its customers' questions
THIS CLIENT'S FILE
PLUMBING CLIENT
Voice profile from its reviews, its front desk, the owner's recorded talk and its customers' questions
Owner edits a draft
THIS CLIENT'S FILE
Stays out of the shared rules
ROOFING CLIENT
Voice profile from its reviews, its front desk, the owner's recorded talk and its customers' questions
THIS CLIENT'S FILE
An agency writing for a niche builds the page map once and reuses it for every contractor. Each contractor's voice and edits stay in that contractor's file, because one owner's taste isn't a law for every other client.

One saved post came from a software agency selling integrated websites to HVAC owners, plumbers and roofers. It was hiring for advertising and lead generation, including copy. Applying the content-and-voice split to its client pages gives the following workflow.

The draft builds one map per page type and reuses it across every client in the niche. A service page gets the same parts each time: who the service is for, the problem in the customer's own words, what the contractor does about it, the proof, the ask. Each part carries its job, so a writer filling in client number seven knows what the proof block has to accomplish before writing a word of it. That shared map is the agency's process, and it's the part that should never be rebuilt per client.

The voice compiler is where the contractors part ways. Each client gets a profile built from its own material: its reviews, the way its staff answer the phone, the owner's recorded talk, the questions its customers ask. Each channel gets its own profile on top. The check is simple: put two client pages side by side, read the openings aloud, and listen for which business is talking.

One rule keeps a system with many clients from drifting. When one contractor edits a draft, that edit goes into that contractor's file. It stays out of the agency's general rules, because one owner's taste isn't a law for every other client. I learned that with the dealer, whose notes are his company's standard and stay there.

A founder's posts go out under the founder's name

A shallow pan sits level at the edge of a dark stream. A few amber nuggets stay in its bottom while grey silt washes back into the current. On a flat rock beside it lie a freshly made amber ring and a small hammer.
A founder's calls supply the facts, opinions and stories worth posting. The spoken wording stays behind, because a phrase pasted from a transcript reads as stitched, and the sentences get written fresh in his measured voice.

The ghostwriting posts were specific about what they feared. One asked for "posts in the founder's voice, not yours," built from founder interviews and sales calls. Another wanted three LinkedIn posts and two X threads a week, and offered a voice pack with his published posts, approved scripts and rejected drafts, each with the reason it was killed. That post said most of the job would be editing and rejecting.

The voice compiler carries the most weight here, and it needs a profile of how the founder sounds when nobody's editing him. Mine was built from about 115,000 words of my transcribed speech, and it measures how often I contract, how long my sentences run before I end them, and which words I never use. A founder's profile gets built the same way, from unscripted calls. His rejected drafts can teach more than his approved ones, because each reason is a rule waiting to be written.

The draft decides what's worth saying this week, and the founder's calls are the best source for it. They supply facts, opinions and stories. They don't supply wording. A spoken phrase pasted into a post reads as stitched: I've seen a post carry "is for light material placement like that," where the "that" pointed three sentences back into a conversation the reader never heard.

Then the founder edits. His pass is a human hand like mine, so the line edit runs again after it, on every sentence he touched, before the post goes out under his name. Every post he kills goes into his file with the reason he gave, and when the same reason shows up twice it becomes a rule.

A brand has a voice guide that hires keep getting wrong

Four lines from a trades branding firm's voice guide, each tagged by who enforces it. No em dashes goes to the checker. Customer framing with no owner blaming goes to the judge. Operator tone goes to the judge and the editor's ear. No one-word sentences goes to the checker.

ONE BRANDING FIRM'S VOICE GUIDE, LINE BY LINE
CHECKER a pattern decidesJUDGE a narrow model call, trained on the firm's approved examplesEAR only the firm's editor can settle it
"No em dashes"CHECKER
Decided every time.
"Customer-framed, never owner-blaming"JUDGE
Needs approved examples of each side before it gets a vote.
"Operator tone"JUDGEEAR
A model flags corporate filler. The editor decides whether it sounds like someone who's run a crew.
"No one-word sentences for punch"CHECKER
Overrides my default, which keeps a deliberate fragment. Recorded in the firm's block.
A voice guide becomes a specification once each line has an enforcer. A pattern decides some lines, a narrowly trained judge decides others, and the firm's editor keeps the call on whether a sentence sounds like someone who's run a crew.

The post closest to the dealer's world came from a branding firm that writes for trades and construction contractors: "Our content lives or dies on voice, and most writers get it wrong on the first try." The firm had already written its voice down as rules: no em dashes; customer-framed rather than owner-blaming; "operator tone," with no corporate fluff or motivational filler; "teaching, not selling"; and flowing prose with no one-word sentences for punch.

The voice compiler's job is to turn each line of the guide into something checkable, and the lines fall into three kinds. "No em dashes" is a pattern; a checker decides it every time. The ban on owner-blaming is a judged rule, and the judge needs examples of each side from the firm's own approved pieces before it gets a vote. "Operator tone" is partly judged and partly ear: a model can flag corporate filler, and only the firm's editor can say whether a sentence sounds like someone who's run a crew.

One line in that guide overrides a default of mine. The line edit protects a deliberate fragment as part of a writer's voice, and this firm bans one-word punch lines and fragment-stacking. The firm wins on its own pages, and its block in the rule list records the exception, so the checker flags those forms on the firm's pages and applies the relevant channel rules elsewhere. When the editor marks a sentence that drifted, the mark goes into the firm's file with the reason, and the guide gets a little more checkable every round.

For a buyer like that, the draft changes little. The firm already knows what each piece is for: every piece teaches one idea an owner can use.

An operator already wired AI into the brand

From above, a heap of violet tiles sits at the top of a sorting table. Four trays below hold them: a large full tray, a smaller one with a brush across it, a few tiles beside a file, and three cracked grey tiles each tied with a blank tag.
One buyer's post asked for someone to decide what stays, what changes, what needs refinement and what never ships. That's a verdict per part, and every draft pulled from the batch carries its reason.

Some buyers are past the question of whether to use AI. One post described a company that had already loaded its brand voice, customer avatars, testimonials and podcast content into its tools, and it was hiring someone to decide "what stays, what changes, what needs refinement, and what should never make it out the door."

That sentence is the draft's verdict step, written by a buyer in his own words. Stays, changes, needs refinement and never ships map onto keep, fix, rework and cut, made one part at a time and recorded with a reason. When one part fails, only that part goes back for rework, and the rest of the batch stays put.

The voice side shows up in a post from the same kind of buyer, hiring for ad scripts narrated by several different people. It asked how a writer keeps each narrator distinct when the drafting tool falls back on the same rhythms every time. That's a voice compiler with several profiles under one brand. Each narrator gets his own: the words he'd use, the length of his sentences, the proof he'd reach for, the jokes he wouldn't make. A script that drifts toward the tool's default rhythm shows up as a distance from that narrator's reference set before anyone records it.

The person signing off on the batch has the job the system can't do without. Whatever that person approves becomes the reference set, and whatever they reject, with the reason, becomes the next rule. What's missing is the record of decisions that turns a reviewer's taste into something the next batch can be checked against.

One brand talks to several audiences on several channels

Two messages from an invented coffee roaster under one band that reads same voice: dry, plain, no exclamation marks. A restock text says the Ethiopia is back from the same farm. A win-back email mentions a changed blend and offers the next bag free.

SAME VOICE Dry, plain, no exclamation marks.
INVENTED EXAMPLE
RESTOCK, TEXT MESSAGE

The Ethiopia's back. Same farm, same roast. It went fast last time.

JOB: Say it's back and get out of the way.
WIN-BACK, EMAIL

It's been a while since your last bag. We changed the house blend this spring and we'd like to know what you think, so the next one's on us.

JOB: Give a lapsed customer a new reason to come back.
An invented coffee roaster with a dry, plain voice sends two messages. The restock text only says the coffee's back, and the win-back email carries a new reason to return.

Lifecycle buyers run flows: a welcome series, a restock notice, a win-back, a review request, each going out by email, text or push to a different group of customers. One retention post put the whole problem in a phrase: "a restock notice reads differently than a win-back." The brand's voice was already documented. The job was making every send sound like that brand while each one did a different job.

The draft asks whether each message earns its place in the journey and what its one job is. A win-back has to give a lapsed customer a reason to come back that wasn't there when they left. A restock notice has to tell someone who wanted a product that it's back, and get out of the way. A message that can't state its job in a sentence is the first cut.

The copywriting pass sets the shape of each series. Every message in a series follows one sequence shape, each message carries one ask, and any loop a message opens gets paid off by a later one. The voice compiler holds the brand voice fixed and moves the tone per channel and per stage.

Here's an invented pair to show it, for a coffee roaster whose voice is dry and plain. The restock text: "The Ethiopia's back. Same farm, same roast. It went fast last time." The win-back email: "It's been a while since your last bag. We changed the house blend this spring and we'd like to know what you think, so the next one's on us." The two share one voice with no exclamation marks.

The customers' own words do a lot of this work. In an older engagement for a supplements brand, the email work mined customer reviews and competitor feedback to write in the language buyers used, and a sequence with no hard pitch ran before the one that asked for the sale.

A content engine has to take positions

Editorial buyers want argument. One post for an interview-based writing role listed what it wanted in plain terms: "Taking a position and defending it. We don't want both-sides hedging." The same post asked for every statistic to be traced to its original source, dated. Several posts in this group also ban generic AI language outright.

Two crossings span a dark gorge. On the left, a short grey plank bridge sags on a thicket of thin sticks. On the right, a bold stone arch leaps across in one span, its end set into a large glowing anchor block bolted into the cliff.
A strong claim backed by a named source can stay strong. Stack 'may,' 'might' and 'can' under an overreach and it still wobbles; name the condition under which it holds and the claim keeps its edge.

Those two demands look like they pull against each other, and the draft is where they stop pulling. The claim checks from the first review do most of the work. A capability doesn't get stated as a guarantee. An anecdote doesn't get stretched into a law. A summary keeps the body's condition. Every claim names who saw it. None of those checks forces a hedge: the first review's advice was to keep a strong claim when the evidence warrants it, and to replace an overreach with the condition under which it holds. The grounding line from the content section is the pattern, a claim kept whole with its condition added.

The copywriting pass then carries the position with proof instead of qualifiers: an exact figure with its source, or a short outside voice with a name on it. A sentence backed that way can keep its strong verb, and that's the version an editorial buyer is paying for.

The voice compiler gives each channel its own editorial voice. A newsletter can carry a longer argument and a dry aside. A LinkedIn post from the same publication makes the claim in its first sentence and the case in three more.

The editor's ear produces some of the most useful data in this setup, and the first review explained why. Rejected suggestions belong in the record with their reasons. If the checker flags a joke it read too literally and the editor keeps the joke, that rejected flag teaches the system something accepted edits can't: where its judgment is wrong.

Four steps. A checker flags the fragments "Evaluating. Refining." The editor keeps them as chosen fragments. The rejected flag goes on record with its reason and licenses the fragment in that client's block. On the next run the checker leaves it alone on that client's pages and still flags it elsewhere.

A REJECTED FLAG IS DATA, SHOWN ON A REAL LINE
FLAG
A checker that wants a verb in every sentence flags Evaluating. Refining.
THE EDITOR
Keeps both: chosen fragments, each with one obvious completion.
THE RECORD
Flag rejected, with its reason. The publisher's block now licenses the fragment.
NEXT RUN
The checker leaves it alone on the publisher's pages and still flags it on the dealer's.
A rejected flag teaches what an accepted edit can't: where the checker's judgment is wrong. Filed with its reason, it becomes a licence in that client's block.

Running the method by hand proved some things and left a gap

In the manual trials described here, the editor's rejected flags were recorded by hand, like nearly every record the method kept. Those trials didn't establish a complete content-and-voice compiler I could hand a client as software. What exists is the method, run on real client work with every operation checked by a person or a second model, to find out whether the quality is reachable at all before anyone builds the software that runs it.

Several parts are proven. The shared minimums can be reached on real volume: on one set of 33 of the dealer's captions, the reviewer's major flags went from 10 to none in one revision, and a later batch went from 8 majors to 3. Per-client rules preserve corrections so later drafts can be checked for the same problem. The publisher's file holds 43 standing rules from eight assignments, each preserving a correction for later drafts to check, and the write passes on the dealer's remaining drafts shrank from 33 captions to 14, then 7, 2 and 1. Drafting first and asking second works too. When a draft needs something only the client knows, it leaves a marked gap, and the client gets at most three questions, each tied to a gap. A marked gap costs one answer. An invented detail costs his trust when he finds it.

In the three-article calibration, the line-edit kit found 84% of the full-book reviewers' edits. Its precision was estimated at about 87%, with the shared-model and reused-example limits described above. Two independent full-book readings put the same four chapters of the craft book on top. And the trace showed where faults enter, which is why every hand gets a second read.

Three columns. Proven: major flags cut from 10 of 33 captions to none, 43 standing rules for one publisher, drafting first with at most three questions, a line edit that found 84% of a full-book review's findings blind, and two reviews agreeing. Not yet proven: whether a voice profile carries the audience's voice, and whether the copywriting pass beats its line-edited version. Still to build: a judge cleared to block, the live-page read, readable profiles, edit capture and the software.

PROVEN
Shared minimums on real volume. Major flags went from 10 of 33 captions to none in one revision, and a later batch from 8 to 3.
Per-client rules stop repeat corrections. One publisher's file holds 43 standing rules from eight assignments.
Draft first, ask second. A marked gap for each missing fact, and at most three questions.
A line edit that holds up blind. It found 84% of what a full-book review found, and about 87% of its own flags matched.
A repeatable reading. Two full-book reviews put the same four chapters on top.
NOT YET PROVEN
Whether a measured voice profile carries the voice an audience hears as one of its own.
Whether the copywriting pass beats its own line-edited version. The blind test is written and hasn't run.
STILL TO BUILD
A judge cleared to block. The 32 judged rules report and warn; none can stop a change yet.
The live-page read after every deploy. It's wired in and skips for want of an access key.
Voice profiles a system can read. Today they're written rules, kept by hand.
Capture of every edit. Today edits get captured and turned into rules by hand.
The content compiler as software a client can log into.
Running the method by hand proved five things and left two questions open. The pattern checks already run on every write and commit; a judge cleared to block, the live-page read, readable profiles, edit capture and the software are still to come.

Two things aren't proven. I don't yet know whether a measured voice profile carries the voice an audience hears as one of its own. The floors measure what's wrong with a sentence, and the voice-fit score measures how close a draft sits to the client's approved pieces. Neither measures whether an operator reading on his phone hears someone who's run a machine. Bell's work on audience design says why that's a separate question: style is a response to an audience, so only the audience can say whether the response landed. And I don't know yet whether the copywriting pass beats its line-edited version: its blind test on three real pieces is written and hasn't run.

So the method treats the audience's ear as data to collect. Every reaction to a real draft gets recorded: what sounded off, in the reader's words, and why. Approved pieces go into the reference set with a note on what makes them work, and the voice profile is designed to be re-derived from that evidence every few dozen records, so it follows what the audience accepts instead of what I guessed it would.

A faint path through a meadow carries two sets of footprints on the same track, then stepping stones laid along it, the newest glowing. Under it, four steps: a client edits a sentence and gives a reason, the edit waits in that client's file, a second edit arrives with the same reason, and it becomes a rule the next draft passes.

HOW A REPEATED CORRECTION BECOMES A RULE
EDIT AND REASON
A client changes one sentence and says why.
CLIENT FILE
The edit waits in that client's file.
SAME REASON AGAIN
A second edit arrives with the same reason.
NEW RULE
The next draft gets checked against it and passes.
Every edit goes into the client's file with its reason. When the same reason shows up a second time, it becomes a rule the next draft has to pass, so nobody makes that correction by hand again.

The ear doesn't have to belong to the client. For the dealer, the ideal reader is an operator who's never seen the brief, reading in the cab, and the owner is the closest reader available. For an editorial buyer, it's the editor's. For a founder, it's his own, because his audience is reading him. Whoever holds the ear, their edits go into the file with a reason, and the profile follows them.

The recorded implementation had gaps between the method and a complete system. The following describes its state during those trials, rather than a current service audit. The minimums live in one list: 55 rules in eight families, with the keep signals and a short block per client. Twenty-one of those rules are patterns a script can decide. They run whenever an agent writes copy into one of my projects and again before every commit, and a hit on a blocking rule stops the change. Thirty-two need a model's judgment. A judge reads for them and reports what it finds, but none of its findings can stop anything yet, because a judged rule gets that power only after a separate evaluator shows it reproduces known answers, and none has been cleared. The last two read the page as it went live, and the job that runs them after each deploy is wired in but, as I write this, skips for want of an access key. Voice profiles exist as written rules per client, kept by hand. Edits get captured by hand and distilled into rules by hand. And the content compiler is a set of operations I run, a model drafting each part and a separate reviewer grading it, and it isn't yet software a client can log into.

Once it's software, it scales the way it's already designed. The writing method lives in one kit per role: the operations, the reference material, the file of corrections and the instructions for each tool. That kit is the same for every client. What changes per client is the voice profile, the reference set and the client's own file of edits. Every client in this article, from the dealer to the AI-tools publisher, would share one writing kit and carry a profile of its own: one toolchain with many targets, applied to voices instead of machines.

What a buyer gets today is the method run on their own work, round by round, with every correction kept and turned into a rule, and the software arriving underneath it as each piece gets built. If you need a tool you can install this week and run without anyone in the loop, this isn't that, and I'll say so on the first call.

Five questions sort a system from a prompt

Put five questions to whoever writes for you, whether that's me, an agency or a tool, and the answers show whether there's a system behind the copy or a prompt and a hopeful reviewer.

A card of five questions for a writing vendor, each with a good and a weak answer: the parts and their jobs, where your voice lives in writing, the minimums every sentence meets, what happens to your edits, and who reads every sentence after the last person edits it.

FIVE QUESTIONS FOR WHOEVER WRITES FOR YOU
ASKA SYSTEM SHOWS YOUA WEAK ANSWER
What are the parts of this piece, and what's each one's job?
A SYSTEM SHOWS YOU
A content map: the parts, their jobs, and which got cut and why.
A WEAK ANSWER
One long draft with no parts named.
Where does my voice live in writing, and can I read it?
A SYSTEM SHOWS YOU
A written voice profile you can open.
A WEAK ANSWER
'In the prompt,' or 'in our writers' heads.'
What minimums does every sentence meet, and how are they checked?
A SYSTEM SHOWS YOU
Named checks, including what catches a claim stronger than its evidence.
A WEAK ANSWER
A banned-word list and nothing else.
What happens to my edits?
A SYSTEM SHOWS YOU
A file of your edits with reasons, turned into rules.
A WEAK ANSWER
Corrections that vanish into the next draft.
Who reads every sentence after the last person edits it?
A SYSTEM SHOWS YOU
A line edit after every human pass, and a read of the live page.
A WEAK ANSWER
Nobody; the editor is the last step.
Ask these of any writer, agency or tool. A system shows you a content map, a written voice profile, named checks, a file of your edits and a read after every hand, and a prompt with a hopeful reviewer can't.
  • What are the parts of this piece, and what's each one's job? A vendor with a content compiler, by any name, can show you the map: the parts, their jobs, and which ones got cut and why.

  • Where does my voice live in writing, and can I read it? If they say "in the prompt" or "in our writers' heads," the voice will drift the first time the writer or the model changes.

  • What minimums does every sentence meet, and how are they checked? A list of banned words is a start. Ask what catches a claim stronger than its evidence, or a summary that drops the body's condition.

  • What happens to my edits? A record of your corrections lets the next writer check for the same problem before sending another draft.

  • Who reads every sentence after the last person edits it? A good answer names a line edit after every human pass, yours and the editor's included, and a read of the page as it went live. If the editor's pass is the last step, the editor's slips ship.

If you publish five pieces a month on one channel and you're happy editing them yourself, a good writer with a good prompt is the right tool, and my original content compiler entry says the same: below a certain volume, the architecture costs more than it saves. If you need a turnkey tool with nobody in the loop, I've told you above where the line sits today. The method is most useful where writing has to keep several clients or channels distinct and repeated corrections already need managing.

The buyers this is for have a different week. They're publishing for many clients, or many audiences, or under someone else's name, and they're fixing the same sentence for the third time. For them, the split makes a correction inspectable: whose voice it belongs to, why the sentence changed, and what the next draft has to get right.

The bolt of amber cloth and three dress forms return with labels. The long overcoat is LinkedIn at 12 words a sentence, the jacket is Facebook at 10, the vest is Instagram at 8, and the stone wall behind them is the shared minimums.

LINKEDINthe long cut: 12 words a sentence
FACEBOOK10 words a sentence
INSTAGRAMthe short cut: 8 words a sentence

Median sentence lengths from one equipment dealer's approved captions.

THE CLOTHone content map
EACH CUTone client, one audience, one channel
THE WALLshared minimums no voice moves
THE SECOND READthe sentence check after every hand
THE EARa reader from the audience decides whether it sounds right
The same cloth is one content map. Each cut fits one client, one audience and one channel, the wall of minimums stays put behind them, every sentence gets a second read after each hand, and a reader from the audience decides whether it sounds right.

Go back to the dealer's posts on three channels. Each post has one map: who it's for, what they get, why they should believe it, what to do next. Above that map sit three channel profiles of one company's voice, each measured from approved captions. Underneath sit shared checks and that client's recorded exceptions, and the sentences get read again after every hand that touches a post. And at the end of every round there's the owner, who told me he wouldn't read for grammar. He'd read for whether it sounds right.

Visuals 24

Keep reading

Case studies from this practice