Voice in writing: how to edit without erasing it

There is a moment in every ghostwriting project where the draft becomes good and the client stops recognising it. The sentences have tightened. The structure is clear. The argument survives a hostile read. And the person whose name goes on the piece reads it twice, says “this is really well written,” and goes quiet in the specific way that means someone else. The piece got better and less like its author at the same time, and the two movements were the same movement.

Everyone asks what is voice in writing when they are in that moment, as if it were a technique they missed in a guide. It is not a technique. Voice is the record of how a particular mind handles disagreement, caveat, emphasis—and it survives editing only when the editing is designed to preserve it. This piece is about that design: what voice actually is on the page, which editorial pass kills it and which one saves it, and how we run a ghostwriting so the finished piece reads like the person signing it wrote it on their best day.

What voice is, operationally

Strip the mysticism and voice in writing is made of three observable things, all of which survive analysis.

The first is position-taking: how fast a writer commits, and how. Some people arrive at a verdict and then qualify it; some hedge their way to a verdict as a way of thinking out loud. Move a paragraph so its caveat comes before the claim instead of after it, and the author will flinch before they can tell you why. That flinch is voice doing its job.

The second is register: what the writer is willing to say bluntly, what they dress up, what they refuse to say at all. A founder who writes “we were wrong about the pricing” and a founder who writes “we identified limitations in our previous model” are describing the same event with different theories about trust. Change one to the other and you have not edited the sentence. You have replaced the person.

The third is rhythm: where they put the short sentence. This one people underestimate because it feels cosmetic. It is not. A long-breathed argument with one five-word sentence dropped into it the way a boxer drops a jab is a signature as distinctive as vocabulary, and it is the first thing a smoothing edit destroys, because flattening rhythm is what competent copy does without anyone deciding to.

Notice that none of the three is vocabulary. People imagine voice as a private dictionary: this person says “look,” that person says “candidly.” Vocabulary is in fact the cheapest and most fakeable layer. A style guide can borrow a lexicon in an afternoon. What cannot be borrowed is a person's way of conceding a point, and that is most of what readers mean when they say a piece “sounds like them.”

Keep the transcript in the room

The defence against erasure is mechanical before it is artistic. Before we draft anything with a named human on it, we interview, and the interview is evidence collection as much as content gathering. While the subject talks, we log the places they return to, the stories they retell unprompted, and, most usefully, the moments they get difficult: how do they introduce a caveat? What do they do when they do not believe the question? Where do they refuse to answer, and is the refusal itself a position worth keeping in the piece?

The log matters more than the recording, and it changes what the draft is allowed to do. When a line in the draft starts drifting toward generic competence, when a sentence could have been written by anyone, we go back to the log and ask how this person would have said it. Sometimes the answer is a phrase they actually used. More often it is a shape: they would have led with the concession, they would have named the number instead of rounding it, they would have been blunter than the draft.

This is also the honest answer to the fear that ghostwriting means surrendering your voice at the door. The risk is real. It is the risk the whole piece is about. But the risk is highest exactly when the writer works from a topic brief and a company deck instead of from recorded evidence of how you think. Our ghostwriting process is built around the interview for that reason. The transcript is the voice's witness stand.

Developmental, copy, line: three passes, three different rights

Most damage is done by editors doing the wrong pass, and it is preventable once you separate the three jobs and, crucially, the rights each one earns.

Developmental editing decides what the piece is: the claim, the structure, what belongs and what does not. It is the pass with the most legitimate authority, because a beautifully voiced argument in the wrong shape still fails. Done well it reads as an argument, not an instruction: “the reader will need this before they believe that”—not “move this section.”

Copy editing fixes what is objectively wrong. Grammar, agreement, a dangling reference, a citation that does not match. What is copy editing, finally, in the narrow sense that matters for voice? A pass that can be checked—there is a right answer, and a second copy editor would make the same change independently. That test is also its licence: changes that pass it can ship silently, because an author who could replicate them cannot reasonably object. Copy-editing marks are not a request for a meeting. A writer who treats every comma as identity politics is asking the reader to trust a draft full of errors.

Line editing is where the war is. What is line editing? Changes to rhythm, word choice, emphasis, order—changes with no objectively right answer, which is exactly why they are load-bearing for voice. Swap “we were wrong” for “our earlier model had limitations” and every grammar checker in the world stays silent, because nothing broke. Something was murdered. When a line edit is done by someone holding only the file, with no interview, no transcript, no standing to be argued with, that is where voice dies: silently, one improvement at a time, until the piece is “really well written.”

Line editing vs copy editing is therefore not a taxonomy quiz. It is an assignment of rights. Senatore runs one rule off that distinction: copy-editing changes ship without comment; line-editing changes carry a reason in the margin, and an unexplained line change counts as a defect in the pass, not in the draft. The reason is not politeness. It is what lets the author object to the judgement instead of squashing the whole revision because they cannot locate the one change that made it sound like nobody.

The style guide trap, and the author's own pass

The editor is not the only threat to voice. Most corporate style guides are voice-erasure machines that somebody once had good reasons for. They begin well: spell out acronyms, serial comma, no exclamation marks. Then, because the writer of the guide could only defend rules they had seen broken, they accrete until they legislate register itself. Never begin a sentence with “but.” Avoid first person singular. Do not use colloquialisms. Each rule is individually sane and together they compile to one instruction: sound like no particular person. A founder who writes “but we were wrong” and hits three rules doing it learns, by the fourth article, to pre-emptively write the way the guide wants, and the erasure is complete before an editor touches a thing. This is why we treat the style guide as an input to the interview, not a replacement for it: the guide says how the company avoids mistakes; the transcript says how the person makes arguments, and the draft needs the second more than the first. Guides can be written the other way. A voice-preserving guide is short, bans things by consequence rather than taste (“no weasel attributions—they cost us credibility with buyers who know the field”), and keeps a short list of house sentences next to its never-list: lines the company likes, kept intact, with a note on why. Rules teach writers what not to do; exemplars teach them what to sound like, and only the second one transfers a voice.

The second erasure is quieter and comes from inside the house: the author's own safety pass. A ghostwritten draft arrives that sounds like the founder having a good week, and the founder, obliged, nervous, or suddenly afraid of their own bluntness, goes through it making it “safer.” The five-word verdict sentence becomes a paragraph of context. The concrete number rounds itself up into “significant.” The concession that made the piece trustworthy gets softened into a boast. Each edit is a small retreat from the position the piece was commissioned to take, and together they return the draft to the competent-voiceless place the project was hired to escape. The tell is that the word count goes up while no paragraph gains content; voice loss is almost always an inflation. The defence costs one sentence at kickoff: the author's edits should hunt for facts, missing cases, and anything the writer could not have known—plus one question, “is there a sentence here you would not say out loud in a board meeting?” If yes, it changes. If no, the urge to soften is a reflex, not a judgement, and naming the reflex before the draft lands is most of the treatment. The ghostwriting engagement keeps the argument the client's from transcript to final change for exactly this reason.

Write long, cut with reasons

The other structural defence is counterintuitive: we draft deliberately long. Not padded—long. The reason is that voice cannot survive a thin draft, because a thin draft leaves no material to cut with judgement, and cuts made for lack of material are indistinguishable from erasure.

Give the editorial pass two live ways of making the same point and the choice between them is an editorial decision, visible and arguable. Give it one way and the editor's only remaining moves are to trim, flatten, and “improve”—the trims that quietly change register, one adverb at a time. Every substantive cut we make gets its reason written down, and the reasons are the deliverable that actually protects voice: they let the author disagree with the judgement rather than the wording, which is the only level at which an author can win an argument with an editor.

Authors can run the same defence at home. Keep your first-draft excess on a separate branch of the document rather than deleting it into the void. Six weeks later, when a reviewer's “can we make this tighter” has quietly amputated the way you put your case, the long draft is the evidence of what you actually sound like.

Before and after: the same sentence, three passes

One worked example from Senatore—anonymised, the kind of line every founder signs off on. The author's transcript shape: blunt verdict, numbers unrounded, mild impatience with vendors. The draft, as it arrived:

We moved our reporting from dashboards to written weekly notes, and honestly the written notes, which force us to actually say what changed and why in prose instead of pointing at a chart and nodding, have, if I'm being completely frank, meaningfully improved the quality of our decisions, and our leadership team has found that they engage far more attentively.

Copy-editing pass (ships silently; every change replicable): comma after “notes,” subject-verb tidied, “completely frank” kept—it is the author's, not an error.

Line edit, bad version: “Our shift from dashboard reporting to written weekly notes has meaningfully improved decision quality and leadership engagement.” Clean. Correct. Publishable by anyone. The author's flinch is guaranteed, because every trace of the impatient human who said “instead of pointing at a chart and nodding” is gone. The concrete image, which was the argument, got sacrificed to a word count.

Line edit, good version, with its margin note: “We replaced dashboards with written weekly notes. Honestly, it worked: you cannot nod at a page of prose, so the notes force us to say what changed and why.” Margin note: split into three beats to keep the verdict first and the image second; cut “completely” from “being completely frank” because the doubled hedge reads as performance—author can push back on the cut in one line.

Same information, same truth, two opposite outcomes. The difference was not talent. It was which pass believed it had the right to touch rhythm and image, and whether it had to account for itself.

The test that settles it

Before anything signed by a human ships, we run one test, and it needs no editor's licence. Read the draft aloud—ideally to the person whose name is on it, and ideally in the room where they would actually say these things. Every sentence they cannot say out loud without wincing goes, no matter how good it looks on the page. The winces are diagnostic: a sentence that reads beautifully and cannot be spoken is written in someone else's voice, usually the editor's, and it is the unit of erasure.

The corollary is the position we hold at the end of every project, and it is the one worth stealing. Polish is not the enemy of voice; silent polish is. If every change to rhythm, word choice, and emphasis has to survive a margin note and a read-aloud, then a piece can get arbitrarily good without ever leaving its author—which is the only standard worth editing to. Better that sounds like you. Anything better is just someone else.

What survives this gauntlet is worth naming, because it is not what most people think voice-preservation costs. A preserved draft is not rougher or less polished than an erased one—it is usually tighter, since the margin notes did the cutting and killed the flailing. The difference is in what the polish was for. Erasure polishes toward the average of everyone who has ever reviewed the document; preservation polishes toward the specific person at their clearest. Only one of those two products can honestly be signed, and readers can feel the difference a page in without explaining it—which is the same sense, running in reverse, that makes a stranger's writing trustworthy. Trust on the page is voice recognised by someone other than its owner, and no style guide has ever produced a drop of it.

If your organisation needs that discipline run on someone else's drafts, the team that applies the rule above is the Imperium programme, and its copy-and-line work is priced on the pricing page; the companion piece on why drafts stall before editing even begins is here, and the enquiry form reaches the editor who will tell you which of the three passes your last project actually needed.

One practical note on running the test with a team, because it fails in meetings more often than it fails on pages. If the answer to the substitution question is argued rather than felt, the room is debating taste and will not settle it; taste arguments end in compromise, and compromise is exactly the erasure the test is built to catch. Run it as a veto instead: any reader on the team may call the piece generic, nobody has to prove it, and the only way to overturn the veto is to read out the specific sentence they believe a stranger could not have written. Sentences survive that reading. Adjectives never do. This asymmetry is why the test works on strangers too, in the wild: the piece a reader forwards is almost never the piece that is most correct, it is the piece with a sentence in it that sounds like a person wrote it, and a person wrote it because somebody in the process refused to let it be smoothed.