# Novel I — Revision: The Line Edit
In ["Novel I — Revision: Structural Edit"](/harrys-desk/novel-i-revision-structural-edit) I said the first pass of revision is architectural: you read the draft cold, map what is on the page, and decide which pillars stay. In ["Novel I — Revision: Scene Deepening"](/harrys-desk/novel-i-revision-scene-deepening) we went deeper — into the four layers that make a scene breathe: stakes, subtext, sensory ground, and the gap between what is said and what is meant. Today we go all the way down. We reach the sentence.
The line edit is the last revision pass before copyediting. It is not about grammar — that comes later, and a copyeditor will catch what you miss. It is about prose quality at the level where prose lives: the word, the clause, the sentence, the paragraph. Rhythm. Precision. The weight of a period. The decision to use a comma or to let the sentence run. The line edit is where you make the writing sound like writing — not like a transcript of thinking, not like a competent summary of a scene, but like sentences a reader will want to read twice.
This is the hardest pass to teach because it is the hardest pass to systematize. Structural editing has rules: if a scene does not advance plot or character, cut it. Scene deepening has diagnostics: check the four layers, fix the missing one. The line edit has no checklist. It has taste. And taste, I will argue, is the one thing you cannot delegate to AI — but you can use AI to sharpen it, if you know what to ask.
What a Line Edit Actually Does
A line edit addresses three things: clarity, rhythm, and resonance. Let me define each precisely, because these words get used loosely and the looseness is where writers lose their footing.
Clarity is whether the sentence means what you intend it to mean. Not whether it is grammatically correct — a sentence can be grammatically impeccable and semantically opaque. Clarity is the reader's experience: did they understand, on first read, without retracing, what the sentence conveyed? If they had to go back, the sentence failed. The line edit catches sentences that are technically sound but cognitively muddy — usually because they are too long, too nested, or because the subject and verb are too far apart.
Rhythm is the musical pattern of the prose — the alternation of long and short sentences, the placement of stressed and unstressed syllables, the pacing of commas and periods. Rhythm is not decoration. It is meaning. A short sentence after three long ones creates emphasis. A long sentence that flows without interruption creates immersion. A sentence that breaks where the emotion breaks creates empathy. When rhythm is wrong, the reader feels it as fatigue or impatience, though they will rarely name it. They will say the writing is "dense" or "slow" or "doesn't flow." What they mean is the rhythm is working against the content.
Resonance is whether the sentence lingers. A clear sentence with good rhythm that says exactly what it means and nothing more is a functional sentence. A resonant sentence does more than function — it opens a door the reader did not know was there. It carries a secondary meaning, an image that echoes, a word choice that implies more than it states. Resonance is what separates prose that communicates from prose that stays with you. It is also the rarest quality and the one most worth editing toward.
The line edit works on all three, in that order. You fix clarity first because unclear prose cannot be evaluated for rhythm. You fix rhythm second because arrhythmic prose cannot be evaluated for resonance. You reach for resonance last, and you do not force it — you recognize it when it is already half-present and you clear away the clutter that obscures it.
The Discipline of the Cut
The single most important skill in a line edit is cutting. Not rewriting — cutting. Most sentences in a first draft are carrying weight they do not need. Adverbs that duplicate the verb. Adjectives that specify what the noun already implies. Transition phrases that signpost what the reader can already see. Hedges — "seemed to," "began to," "almost" — that soften moments which should not be softened.
I want to be specific about what cutting means because writers hear "cut" and think "remove." Cutting is not removal. It is concentration. When you cut a word from a sentence, the sentence does not become less. It becomes more. The remaining words carry the meaning the cut word was diluting. The sentence gets denser, sharper, closer to what you actually meant.
Consider this sentence from a hypothetical draft:
> She slowly began to walk across the room, her footsteps echoing softly against the cold stone floor.
Nineteen words. What is actually happening? A woman walks across a stone floor. The rest is padding. "Slowly" is redundant with "began to" — both signal a lack of urgency, and the doubling kills both. "Softly" is redundant with the context — footsteps on stone echo; "softly" tries to modify the echo while the sentence structure says the footsteps are echoing, which is a contradiction. "Cold" is a sensory addition, but it is generic — every stone floor in every draft is cold. The sentence as written is a first-draft sentence: it conveys the event but dilutes it with qualifications that cancel each other.
The line-edited version:
> Her footsteps echoed across the stone floor.
Seven words. The woman is walking — "footsteps" implies motion. The echo is there. The stone is there. The cold is gone, but the cold was not earning its place. What remains is a sentence that does one thing clearly and has rhythm — four beats with a natural pause after "echoed." The reader sees the room. The reader hears the sound. The sentence is over before it wears out its welcome.
This is the line edit in miniature. You are not adding. You are removing what does not belong so that what does belong can breathe.
Where AI Helps and Where It Harms
The line edit is where AI is most useful and most dangerous, and the difference depends entirely on how you deploy it. Let me be precise.
AI is excellent at flagging clarity problems. Paste a paragraph and ask: "Which sentences are difficult to parse on first read? Identify structural problems — nested clauses, long subjects, delayed verbs — but do not rewrite." The model will find your muddy sentences with reasonable accuracy. It is pattern-matching against millions of well-formed sentences, and when one of yours deviates from the pattern in a way that impairs readability, it notices. This is useful. It is the same thing a copyeditor does in the margin with a query mark.
AI is decent at rhythm diagnosis. Paste a paragraph and ask: "Analyze the sentence length variation. Is there a monotony problem — too many sentences of similar length? Are there sentences that should be shorter for emphasis?" The model can count sentence lengths and identify runs. It will not always be right about which sentences should be shorter, because it does not feel the rhythm the way a reader does. But it will show you the pattern, and seeing the pattern is half the work.
AI is bad at resonance. This is where the trouble starts. If you ask a model to improve a sentence, it will usually make it clearer and more rhythmic and less resonant. The model optimizes for the first two qualities — clarity and rhythm are patterns it can match — but resonance is idiosyncratic. It depends on context the model does not have: what came before, what the character is feeling, what word will echo three pages later. The model does not know the book. It knows the sentence. And a sentence optimized in isolation is a sentence stripped of its relationship to everything around it.
Here is the rule I follow, and I recommend it without qualification: use AI to identify problems. Fix them yourself.
When the model tells me a sentence is unclear, I rewrite it. When the model tells me my sentences are all the same length, I vary them. When the model flags a word as repeated, I decide whether the repetition is intentional — sometimes it is, and the model's flag is wrong because repetition can be resonance. But I never paste a paragraph and ask the model to improve it. That produces prose that is smoother and less alive. The model's version of my sentence is always more correct and less mine.
The Three-Pass Method
I line-edit in three passes, each with a single focus. Trying to fix everything at once leads to fixing nothing well.
Pass One: Clarity
I read the chapter at normal speed. Every time I stumble — every time I have to slow down, reread, or parse — I mark the sentence. I do not fix it yet. I am mapping the trouble spots. When I have finished the chapter, I go back to each marked sentence and ask: what is wrong here? Usually it is one of three things. The sentence is too long and loses its thread. The subject and verb are separated by too many clauses. Or the sentence is trying to do two things at once and should be split.
The fix is almost always simplification. Shorten the sentence. Move the subject and verb closer together. Split the double-duty sentence into two single-duty sentences. I am not changing the meaning. I am making the meaning accessible.
Pass Two: Rhythm
I read the chapter aloud. Not whispering — aloud, at speaking volume, at the pace I would read to an audience. This is the single most valuable technique in the line edit, and almost no one does it. Reading aloud forces you to experience the prose as sound rather than as information. Every sentence that makes you trip, every clause that runs out of breath, every paragraph that drones — these are rhythm failures.
The fixes are structural. Break long sentences at natural pause points. Combine choppy short sentences into a flowing one. Move a short sentence to the end of a paragraph for emphasis. Vary the opening structures — if every sentence begins with the subject, the rhythm goes flat. Sometimes the fix is a single comma. Sometimes it is reordering clauses. The goal is not musical perfection. The goal is to make the prose sound like it has a pulse.
I do not use AI for this pass. Reading aloud is something the model cannot do for me, and the physical experience of speaking the prose teaches me things the model never will. If you skip every other technique in this article, do not skip this one. Read your chapters aloud. The rhythm problems will announce themselves.
Pass Three: Resonance
This is the pass where I am looking for the moments that are already almost there — the sentences that have a flicker of something beyond function — and clearing away what obscures them. I am not adding resonance. I am uncovering it.
A sentence in the draft reads: "She stood at the window and watched the rain, thinking about what her mother had said, feeling the weight of the words settling over her like something she could not shake." That is a first-draft sentence trying to be resonant and failing because it explains the resonance instead of creating it. "Feeling the weight of the words settling over her like something she could not shake" is the writer telling the reader what to feel. The resonance is in "the weight of the words." Everything after that is explanation.
The line-edited version: "She stood at the window. The rain. Her mother's words, settling." Three sentences. Fragmentary. The rhythm breaks — which is what the character is experiencing. The resonance is in the space between the sentences, in what the reader supplies. "Settling" does the work of "like something she could not shake" without telling the reader how to feel. The reader feels it because the sentence structure creates the sensation of weight descending and then stopping.
This kind of edit is the hardest to teach because it requires you to feel the sentence as a reader, not as a writer. You have to read it and ask: does this make me feel something, or does it tell me to feel something? If it tells, it is not resonant. If it makes, you leave it alone. If it is almost making — if the raw material is there but the scaffolding is showing — you cut the scaffolding and let the material stand.
The Paragraph as a Unit
Most line-editing advice focuses on sentences. I want to add a word about paragraphs, because paragraphs are where sentence-level craft becomes structural.
A paragraph is not a container for sentences. It is a unit of attention. When the reader reaches a paragraph break, they take a micro-breath. The break signals: a new thought, a shift in time, a change of camera. If your paragraphs are all the same length, the reader's attention goes flat. If your paragraphs are all long, the reader gets tired. If they are all short, the reader gets jittery.
In the line edit, I look at paragraph length the way I look at sentence length. I want variation. A four-line paragraph followed by a one-line paragraph followed by an eight-line paragraph creates a rhythm at the macro level that mirrors the sentence-level rhythm. The one-line paragraph hits like a drumbeat. The eight-line paragraph creates immersion. The four-line paragraph is the baseline.
I also look at where paragraphs end. The last sentence of a paragraph carries more weight than the sentences before it — it is what the reader holds as they pause. If the last sentence is weak, the paragraph deflates. If the last sentence is strong, the paragraph reverberates. I will often move sentences within a paragraph so the strongest one is last. This is a small change with a large effect.
The Line Edit and Voice
One final consideration. The line edit is where voice lives — or dies. Every cut, every rhythm change, every word substitution is a voice decision. When you change "walked" to "moved," you have changed the voice. When you break a long sentence into two short ones, you have changed the voice. When you cut an adverb, you have changed the voice.
This is why you cannot delegate the line edit to AI. The model does not know your voice. It knows an average of all voices, which is no voice at all. If you let the model line-edit your prose, you will get prose that is clearer, more rhythmic, and fundamentally anonymous. It will read like competently edited prose by no one in particular.
Your voice is in your choices — the words you reach for, the rhythms you prefer, the sentences you let run and the ones you cut short. The line edit is where you make those choices deliberately, with full attention, one sentence at a time. It is slow. It is unglamorous. It is the difference between a draft that is finished and a book that is written.
For Next Time
On Monday we begin Novel II — Deep Craft, starting with ["Novel II — Deep Craft: Point of View"](/harrys-desk/novel-ii-deep-craft-point-of-view). We will leave the revision passes behind and enter the territory of advanced technique: how POV choice shapes everything from sentence structure to reader trust, how to manage distance between narrator and character, and how to use AI to test whether your POV is consistent without letting it flatten the particularity of a close first-person or the authority of a wide omniscient.
Your homework until then: take one page — just one — from your current draft. Read it aloud. Mark every sentence where you stumble. Then take those sentences and cut them by a third. Not rewrite — cut. Remove words until the sentence does its work and nothing more. When you have finished, read the page aloud again. If it sounds better, you have just done a line edit. If it sounds worse, you cut something that was carrying weight you did not recognize — put it back and try a different cut. The goal is not perfection. The goal is to develop the ear that hears the difference between a sentence that works and a sentence that sings.
---
*Harry Mercury, Editor in Chief* *The SMF Works Project* *Week 13, Article 3*