--- title: "The Developmental Edit Nobody Warns You About (And How AI Catches What You Miss)" url: https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo canonical: https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo language: en published: 2026-10-10T17:00:00.093Z category: "AI for Creators" author: "Mac" summary: "Elena finished her first draft on a Sunday night in March. Eighty-six thousand words. She printed nothing, told no one, and went to bed feeling like a person who had done something difficult..." html_alternate: https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo llm_index: https://lacunai.com/blog/llms.txt atom_feed: https://lacunai.com/blog/feed.xml --- # The Developmental Edit Nobody Warns You About (And How AI Catches What You Miss) > Elena finished her first draft on a Sunday night in March. Eighty-six thousand words. She printed nothing, told no one, and went to bed feeling like a person who had done something difficult... ## Table of contents - [What a Developmental Edit Actually Is (and What It Is Not)](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#what-a-developmental-edit-actually-is-and-what-it-is-not) - [The cost problem](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-cost-problem) - [The Skepticism, Addressed Up Front](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-skepticism-addressed-up-front) - [The honest limits](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-honest-limits) - [Why You Cannot Edit Your Own Draft by Reading It](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#why-you-cannot-edit-your-own-draft-by-reading-it) - [The Four Layers of a Useful Diagnostic](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-four-layers-of-a-useful-diagnostic) - [Layer One: Scene Function](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#layer-one-scene-function) - [A worked example](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#a-worked-example) - [The scene function checklist](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-scene-function-checklist) - [Layer Two: Scene Velocity and Manuscript Pacing Issues](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#layer-two-scene-velocity-and-manuscript-pacing-issues) - [Three kinds of drag](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#three-kinds-of-drag) - [The velocity chart](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-velocity-chart) - [Pacing is relative to genre](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#pacing-is-relative-to-genre) - [Layer Three: The Emotional Arc](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#layer-three-the-emotional-arc) - [What a healthy arc looks like](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#what-a-healthy-arc-looks-like) - [Mapping emotion honestly](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#mapping-emotion-honestly) - [Layer Four: Stakes, Tension, and the Villain Problem](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#layer-four-stakes-tension-and-the-villain-problem) - [The antagonist who wants nothing](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-antagonist-who-wants-nothing) - [The stakes ladder](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-stakes-ladder) - [The Self-Editing Novel Checklist](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-self-editing-novel-checklist) - [Pass one: the inventory](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#pass-one-the-inventory) - [Pass two: the function audit](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#pass-two-the-function-audit) - [Pass three: the velocity and arc review](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#pass-three-the-velocity-and-arc-review) - [Pass four: the stakes interrogation](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#pass-four-the-stakes-interrogation) - [A quick reference](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#a-quick-reference) - [How to Actually Prompt AI Developmental Editing for Fiction Diagnostics](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#how-to-actually-prompt-ai-developmental-editing-for-fiction-diagnostics) - [Working at manuscript scale](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#working-at-manuscript-scale) - [The Mistakes Writers Make With Developmental Editing Tools](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#the-mistakes-writers-make-with-developmental-editing-tools) - [Accepting every note](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#accepting-every-note) - [Rejecting every note](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#rejecting-every-note) - [Using it to avoid the hard decision](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#using-it-to-avoid-the-hard-decision) - [Skipping the human altogether](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#skipping-the-human-altogether) - [Revising in the wrong order](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#revising-in-the-wrong-order) - [What Happens When You Use This Before You Hire an Editor](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#what-happens-when-you-use-this-before-you-hire-an-editor) - [A Few Honest Words on Craft and Trust](https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo#a-few-honest-words-on-craft-and-trust) Elena finished her first draft on a Sunday night in March. Eighty-six thousand words. She printed nothing, told no one, and went to bed feeling like a person who had done something difficult and permanent. By Thursday she had sent it to three friends. By the following Tuesday she had three kinds of response: "I loved it," "I loved it, but the middle was a little slow," and a long, kind, circling message from her sister that never once used the word "problem" and somehow described nothing else. None of it was useful. That is not an insult to her friends. They did what friends do. They read as readers, they felt something at the end, and they reported the feeling. What they could not do was tell Elena that her protagonist stops wanting anything between chapters nine and fourteen, or that the villain's plan is announced twice and demonstrated never. Nobody warned her this was coming. Nobody warns anyone. This is the developmental edit nobody warns you about. It does not arrive as a red pen on a sentence. It arrives as a question about the whole book: what is this, and does it work? And for most writers, the first time they hear that question asked properly is after they have already paid for the answer, or after an agent has declined to ask it at all. This essay is about getting that question asked earlier, cheaper, and more systematically. It is about AI developmental editing for fiction, and I want to say at the start exactly what I mean by that phrase, because I expect you are already skeptical. (Good. You should be.) ## What a Developmental Edit Actually Is (and What It Is Not) Let us get the vocabulary straight, because the confusion here costs writers real money. A line edit works at the level of the sentence. It asks whether this paragraph is clear, whether this verb is doing its job, whether you have used "suddenly" four times on one page. A copyedit works at the level of correctness: grammar, consistency, house style, the thing the Chicago Manual of Style exists to settle. A proofread catches what survives both. A developmental edit works at the level of the book. It asks about structure, character, pacing, stakes, point of view, and promise. It is the only kind of edit that can tell you the problem is not your prose. The problem is that chapter eight does nothing, and no amount of polishing will make a scene that does nothing do something. The order matters. Developmental first, then line, then copy, then proof. Writers who reverse it end up sanding the legs of a table that is going in the fire. I have watched people spend four months perfecting the opening chapter of a book whose real beginning was chapter three. The prose was lovely. It was also in the wrong place. ### The cost problem Here is the practical difficulty. Developmental editing is skilled, slow work, and it is priced accordingly. The Editorial Freelancers Association publishes a rate chart that is worth reading once, if only to understand why the numbers are what they are. Developmental editing of a full novel commonly lands in the range of a few thousand dollars per pass. A first draft rarely survives one pass. Many need two or three. That is not a scandal. A good developmental editor reads your whole manuscript, thinks about it for hours, and writes an editorial letter that can change the shape of your book. The price reflects the labor. But it also means that a writer on a first draft faces an unpleasant choice. Pay several thousand dollars to learn that the book has structural problems you could have spotted yourself, or skip the edit and submit a manuscript carrying those problems into the world. Neither is good. There is a third option, and it is the subject of this essay: do the diagnostic work yourself, with better instruments, before you spend a cent on a human editor. Then the money you do spend goes toward the hard questions rather than the findable ones. ## The Skepticism, Addressed Up Front I would like you not to throw things. When writers hear "AI" and "editing" in the same sentence, they picture one of two things. Either a button that rewrites your chapters in a smooth, flavorless, committee-approved voice. Or a chatbot that tells you everything is wonderful and your protagonist is "so relatable." Both fears are reasonable. Both describe real failures. And neither is what I am proposing. So let me say plainly what AI developmental editing for fiction is not. It is not rewriting. If a tool is generating replacement prose for your scenes, it is doing a different job, and you should decide deliberately whether you want that job done. This essay is not about that job. It is not a verdict. A model cannot tell you whether your book is good. It has no taste to consult, no market instinct, no ten years of watching what sells. Anyone who tells you otherwise is selling something. It is not a substitute for a human editor. It is a way to arrive at the human editor with a cleaner manuscript and sharper questions. What it is, then, is diagnostic. Think of it as an objective reader with a perfect memory and no social obligations. It will read eighty thousand words in one sitting and hold all of them in view at once, which no human reader can do. It will not get tired in chapter eleven. It will not soften a note because you seem fragile. It will not remember that it liked your opening and let that color its view of your ending. That is a narrow capability. It is also an enormously useful one, because the problems a developmental edit finds are mostly problems of pattern across distance. The villain who vanishes for forty pages. The subplot that gets set up in chapter four and never returns. The emotional beat that lands in act one and then gets repeated, identically, in act three. You cannot see these from inside the draft. You are too close, and your memory of the book is a memory of your intentions, not of the pages. ### The honest limits Notice what I am doing here. I am telling you what the tool cannot do before I tell you what it can. I do this on purpose. A diagnostic instrument is only as good as your understanding of its blind spots, and you will use it better if you know where it goes dark. It will misread irony. It will sometimes flag a deliberate slow burn as a pacing failure. It will occasionally invent a plot hole that is really a thing you handled gracefully on page two hundred. It does not know your genre's unwritten contract unless you tell it. And it can be confidently wrong, in the smooth, fluent tone that makes wrong answers feel right. So the rule is simple. Treat every finding as a hypothesis, not a diagnosis. The tool points. You look. If the thing it pointed at is really there, you fix it. If not, you move on, and you have lost nothing but a minute. That asymmetry is the whole value proposition. A false alarm costs you sixty seconds. A missed structural problem costs you an agent's patience or a reader's trust. ## Why You Cannot Edit Your Own Draft by Reading It Before we get to method, it is worth understanding why the problem is hard. Writers are told to "put the draft in a drawer and come back with fresh eyes," and that is fine advice as far as it goes. It does not go far enough. Here is what happens when you reread your own manuscript. You do not read the words on the page. You read the words and the book in your head, simultaneously, and you cannot separate them. When chapter eight has a gap in motivation, your mind fills it. You know why Mara goes to the harbor. You wrote the reason in a notebook in January. The reader has no notebook. This is the curse of the author's knowledge, and it is not a character flaw. It is how minds work. You cannot unknow your own book. The usual remedies are beta readers and editors, both of which work because they are other people with other heads. The AI diagnostic works for the same reason, with one difference. A beta reader gives you a feeling and sometimes a reason. A model, used well, gives you a structured account of what is on the page, scene by scene, which you can then compare against what you meant to put there. The gap between those two things is where your revision lives. If you have already built a story bible, you have a head start. I wrote about this in The Broken Chapter Four: Why Outlines Collapse and How Story Bibles Save Them, and the principle carries straight over. A bible records what you intended. A diagnostic pass records what you executed. Lay them side by side and the discrepancies announce themselves. If you do not have a bible, you can build a rough one from the diagnostic output itself, which is a good use of the tool. ## The Four Layers of a Useful Diagnostic Good diagnostic work looks at four things. I will name them now and then spend the rest of the essay on how to interrogate each one. One: scene function. What is each scene for? Two: scene velocity. How fast does each scene move, and does that speed match what the story needs? Three: emotional arc. What does the reader feel, and how does that change across the book? Four: stakes and tension. What can be lost, by whom, and how clearly does the reader know? Most developmental problems live in one of these four. Learn to ask about all of them, in order, and you will catch the large majority of what an editor would catch. ## Layer One: Scene Function The most useful single question in all of developmental editing is this: what changes in this scene? Not "what happens." Plenty happens in plenty of scenes that do nothing. A character drives to a restaurant, orders a meal, makes small talk, and drives home. Things happen. Nothing changes. A scene earns its place when something at the end is different from the beginning: a decision made, a relationship shifted, a piece of information revealed, a stake raised. This is where the diagnostic starts, and it is where AI is genuinely good, because the task is tedious for a human and trivial for a model. Feed the manuscript in scene by scene and ask for a plain-language summary of three things: the situation at the start, the situation at the end, and the difference between them. Do not ask for opinions yet. Ask for facts. What you get back is a table of your own book, written by a reader who has no stake in flattering you. Read it slowly. Scenes where the "difference" column is empty or vague are your first suspects. ### A worked example Elena ran her draft through this process. Her novel follows a lighthouse keeper's daughter, Mara, who discovers that the shipping company funding the town is smuggling something it should not. Eighty-six thousand words, thirty-eight scenes. The scene-function pass flagged six scenes where the change column read some version of "Mara feels uneasy" or "Mara thinks about what she has learned." Reflection, in other words. Six scenes in the middle third of the book where Mara received no new information, made no choice, and was not opposed by anyone. Elena's first reaction was defensive. Those were the quiet scenes. Those were the character moments. Readers need room to breathe. She is not wrong that readers need room to breathe. This is the "permission with caveat" moment: yes, you may have quiet scenes. But a quiet scene still has to do something. It can deepen a relationship, plant a clue, or tighten a pressure. The six flagged scenes did none of these. They were pleasant. They were also, as it turned out, the exact pages her sister's email had circled without naming. Elena cut two outright, folded the useful half-paragraph from a third into a neighboring scene, and rewrote the remaining three so that each one ended with Mara knowing, or choosing, something she had not before. The book lost four thousand words and gained momentum. That is what a good developmental pass looks like. It is rarely dramatic. It is usually a matter of asking one honest question thirty-eight times. ### The scene function checklist Use these prompts, in this order, for every scene. Ask yourself three questions. One: what does the point-of-view character want in this scene? Two: what stands in the way? Three: what is different when the scene ends? - Can you state the scene's purpose in one sentence without using the word "and"? - If you deleted the scene, would any later scene stop making sense? - Does the scene end in a different place, emotionally or informationally, than it began? - Is there a choice, a revelation, or a reversal? (One is enough. Zero is a problem.) - Does the scene do more than one job, and if so, is each job earning its space? The second question is the sharpest. If you can delete a scene and nothing downstream breaks, the scene is decoration. Decoration is allowed in small amounts. It is not allowed to make up a fifth of the book. ## Layer Two: Scene Velocity and Manuscript Pacing Issues Now the thing everyone means when they say a book "drags." Writers use pacing as a catch-all, and it is worth separating its parts, because manuscript pacing issues have more than one cause and each cause has a different cure. I use the term scene velocity for how quickly a scene moves the story forward relative to the space it occupies. A two-page scene that reveals the villain's identity has high velocity. A ten-page scene in which two characters discuss the weather has low velocity. Velocity is not the same as action. A quiet conversation can be very fast if every line shifts the balance between the speakers. A car chase can be very slow if nothing about it matters. ### Three kinds of drag When a section drags, one of three things is usually going on. The first is low density. The scene is too long for the amount of story it carries. Fix: compress. Cut the throat-clearing at the front, the lingering at the back, and the dialogue that exists only to be polite. The second is low pressure. The scene is the right length but nothing is at risk. The character is not threatened, the goal is not contested, the reader does not worry. Fix: raise the stakes or add opposition. This is a structural repair, not a trim. The third is repetition. The scene does something the book has already done. A second argument that plays the same beat as the first. A third training montage. Fix: cut one, or make the second version escalate rather than echo. A model can help you tell these apart. Ask it to measure each scene's word count against the amount of new information, new choices, and new conflict it introduces. Scenes with a high word count and a low count of new elements are the drag. Then ask which of the three kinds it looks like. The tool will not always be right. It will be right often enough to save you a week of staring. ### The velocity chart Tracking narrative velocity scene by scene: comparing page count against new information and dramatic pressure. Here is the exercise I would most like you to try. Build a simple chart, one row per scene, with these columns: scene number, word count, point-of-view character, what changes, and a velocity rating from one to five. You can get the first four from the diagnostic pass. The fifth is a judgment call, and I want you to make it yourself, then compare it to what the model says. When you plot velocity across the whole manuscript, patterns jump out. A healthy book is not uniformly fast. It breathes: a sprint, a recovery, a sprint, a longer recovery, a final push. But the breaths have a rhythm, and rhythm can be seen on a chart long before it can be felt on the page. The failure shapes are easy to spot once you know them. A long flat stretch in the middle, five or six consecutive scenes all rated two. A front-loaded book that opens at a four and slowly decays. A book that sprints for the first half, stalls, and then sprints again, with nothing connecting the two bursts. The sagging middle is the most famous, and for good reason: it is where a draft's momentum most often runs out, because the opening was powered by the premise and the ending is powered by the climax, and the middle has only the plot to rely on. Elena's chart showed exactly that. Scenes fourteen through twenty-one sat in a low plateau. Same six reflective scenes, plus two long travel sequences. She had felt the sag without seeing it. The chart made it a fact rather than a mood. ### Pacing is relative to genre One caution, and it matters. There is no universal right pace. A literary novel can sustain a velocity that would kill a thriller. A cozy mystery can linger over tea in ways a hard-boiled detective story cannot. If you hand a model your quiet, interior, slow-building literary novel and ask whether it is paced like a page-turner, it will say no, and it will be correct, and it will be irrelevant. So tell the tool what you are writing. State the genre, the comparable titles, and the reader's expected experience. Then ask it to evaluate pacing against that contract. A book is paced well when it keeps the promise it made to its own audience. The question is never "is this fast?" The question is "is this as fast as this book needs to be?" Jane Friedman has written usefully on the key elements of a book that readers and editors weigh, and her framing of genre expectation is worth keeping in view while you read any diagnostic output. The tool can chart velocity. Only you know what velocity your book promised. ## Layer Three: The Emotional Arc This is the layer writers forget, and it is the one readers remember. People do not finish a novel and recall the plot in order. They recall how it felt: the dread in the middle, the grief near the end, the relief of a particular reunion. That feeling has a shape across the book, and the shape can be mapped. An emotional arc is the trajectory of what the reader, not necessarily the character, is feeling from scene to scene. A model can chart this reasonably well. Ask it, for each scene, to name the dominant emotion a reader is likely to experience, and to rate its intensity on a scale. Then plot the result. ### What a healthy arc looks like Do not expect a smooth upward line. Good emotional structure oscillates. Tension builds, releases, builds higher. Hope rises and gets undercut. A moment of warmth sits just before a drop. The oscillation is the point: contrast is what makes both the highs and the lows register. What you are looking for in your own chart are the failure modes. The flatline. A stretch of scenes where the emotional register never changes. Ten scenes of mild worry. Fifteen pages of vague melancholy. The reader's attention drifts because nothing is asking them to feel anything new. The premature peak. The strongest emotional moment arrives at the seventy percent mark and the rest of the book is a long coast downhill. The climax has nowhere to go because you spent it early. The unearned spike. A huge emotional beat, a death or a betrayal or a reconciliation, that arrives without enough setup. The model can sometimes catch this by noticing that the reader has been given no reason to care about the character or relationship involved. The repeated note. The same emotion, in the same register, struck again and again. Four scenes of Mara's guilt, all identical. By the third, the reader has stopped feeling it and started skimming. Elena's arc chart was flat in the middle, which matched her velocity chart, which matched the six reflective scenes. Three separate instruments pointing at the same forty pages. That convergence is the signal. When more than one lens flags the same region, you are not looking at a quirk of the tool. You are looking at the real problem. ### Mapping emotion honestly A caveat on method. The tool is predicting a reader's likely response, not measuring one. It is working from patterns in how text tends to affect people. That makes it decent at gross shapes and unreliable at fine distinctions. It can tell you a scene is tense. It is less able to tell you whether the tension is the exciting kind or the tedious kind. So use the emotional chart to find regions, not to adjudicate scenes. When it shows you a flat stretch, go read that stretch yourself, with the chart beside you, and ask whether the flatness is real. Often it is. Occasionally you will discover that what the model read as flat was actually a deliberate calm before a storm, in which case the chart has done its job by making you check. ## Layer Four: Stakes, Tension, and the Villain Problem Now the question that opens most editorial letters, and the one that usually hurts: what is at stake? Stakes are what the protagonist, and by extension the reader, stands to lose. Without them, there is no tension, because tension is the reader's awareness that something could go wrong. A story with no clear stakes can be beautifully written and still read like a pleasant walk. A story with clear stakes can be clumsy and still compel. The premise of this essay mentioned an antagonist who lacks clear stakes. I want to unpack that, because it is more common than writers expect and it is hard to see from inside. ### The antagonist who wants nothing Here is how it usually goes. The writer builds a villain, gives them menace, gives them a plan, and then forgets to make the plan cost anything. The villain is threatening in the abstract. They may do something terrible. But the reader never understands what the villain wants specifically, what they will do if they succeed, or what happens to the protagonist in that case. The threat is a mood rather than a mechanism. A diagnostic pass catches this by asking directly. For each scene featuring or referencing the antagonist, extract three items: what the antagonist wants in this scene, what they have done or will do to get it, and what that means for the protagonist if it works. If the answers get vaguer as the book goes on, or if the third item is empty, you have found your problem. In Elena's book, the shipping company was supposedly the villain. The tool's summary of the antagonist's role in the first half read, in effect: "The company is described as secretive and powerful. Its specific goal is not stated. The consequences for Mara if the company succeeds are not stated." That was true. Elena knew exactly what the company was smuggling and why. She had simply never put it on the page in a way that tied it to Mara's life. The fix was not to write more menace. It was to make the cost specific and personal. The company's success would mean the lighthouse closing, which would mean Mara's father losing the only work he had ever done and the house that came with it. Now the villain's plan threatened something Mara could name. Every scene touching the company suddenly carried weight, because the reader knew what it was weighing against. ### The stakes ladder Good stakes escalate. They do not stay at one level for four hundred pages. A useful model is a ladder with three rungs: personal, relational, and larger. Personal stakes are what the protagonist stands to lose directly: safety, a goal, a belief about themselves. Relational stakes involve the people they love: a friendship, a family, a trust. Larger stakes reach beyond them: a community, a principle, a way of life. The strongest books touch all three, and they climb. Early on, the stakes are personal. By the middle, the relational ones join. By the climax, the larger ones are in play, and the protagonist must choose among them. Ask the model to chart where your book sits on this ladder, act by act. If it shows you stuck on one rung, you have a diagnosis. A book that stays entirely at the personal level can feel small. A book that jumps to the larger stakes without grounding them in anything personal can feel abstract and cold. ## The Self-Editing Novel Checklist The four-pass diagnostic audit: interrogating scene function, pacing velocity, and character stakes systematically. You now have four layers. Here is how to put them into practice as a repeatable process, a self-editing novel checklist you can run on any draft. It works in four passes, and I would like you to do them in this order. ### Pass one: the inventory Break the manuscript into scenes. For each, record the point-of-view character, the location, the word count, and a one-line summary of what happens. A model can do the first draft of this in minutes. You should then read it and correct it, because summarizing is where the tool most often misreads. When the inventory is done, read the summaries top to bottom, without looking at the manuscript. You are reading the skeleton of your book. Does it hold together? Does one scene lead to the next? Are there places where you would ask "wait, why did that happen?" If so, mark them. ### Pass two: the function audit For each scene, fill in the change column. What is different at the end? Be strict. "Mara learns more about the company" is not a change. "Mara learns that the company has been bribing the harbor master" is. Specific beats vague, every time. Flag every scene with no change, a vague change, or a change that merely repeats an earlier one. Then decide, scene by scene, whether to cut, merge, or rewrite. A useful rule: if a scene has no change and no plant, it goes. If it has a plant but no change, see whether the plant can live somewhere else. ### Pass three: the velocity and arc review Build the velocity chart and the emotional arc chart. Lay them side by side. Look for convergence: regions where both are low or flat. Those are your sag zones. Then look for divergence: a region with high velocity and flat emotion usually means plenty of action that the reader does not care about. A region with low velocity and rising emotion is often a quiet scene doing its job. For each flagged region, diagnose using the three kinds of drag. Low density, low pressure, or repetition. Pick the cure that matches. ### Pass four: the stakes interrogation For the protagonist and the antagonist, extract wants, obstacles, and consequences across the whole book. Chart them against the stakes ladder. Look for the vanishing antagonist, the vague threat, and the stakes that never climb. Then ask the hardest question on the list: if the protagonist fails, what exactly happens, and does the reader know it by the end of act one? If the answer is fuzzy, that is the first thing to fix, because everything else sits on top of it. ### A quick reference - Scene function: every scene changes something specific. - Velocity: no more than two or three consecutive low-rated scenes without a reason. - Emotional arc: contrast and escalation, no long flatlines, no premature peak. - Stakes: clear by the end of act one, climbing across the book. - Antagonist: a specific want, a specific method, a specific cost to the protagonist. That is the whole checklist. It is not exotic. Every item on it appears, in some form, in the structural frameworks writers have used for decades, whether you prefer three acts, the beat-sheet tradition, or the more analytical approach of Dramatica's theory of story, which breaks a narrative into precise structural components. The point is not to pick the right framework. The point is that every serious framework asks versions of the same questions, and a model can help you ask them across the whole manuscript instead of one chapter at a time. ## How to Actually Prompt AI Developmental Editing for Fiction Diagnostics The quality of what you get depends on the quality of what you ask. Vague prompts get vague answers. "Tell me what you think of my novel" will return something warm and useless. I want you to write prompts that behave like an editorial checklist. A few principles. Ask for evidence. Tell the tool to quote or cite the specific passage that supports each finding. This does two things. It forces the model to ground its claims in the text, and it lets you verify them in seconds. If it cannot cite a passage, you should be suspicious of the claim. Separate observation from judgment. First prompt: summarize, extract, chart. Second prompt, later: evaluate. If you mix them, the model will start praising or criticizing before it has finished describing, and you will not be able to tell which statements are facts. State the contract. Tell it the genre, the target reader, the comparable books, and what you intend each act to accomplish. A diagnostic without a standard is just an opinion. Ask it to argue against itself. After a finding, ask: "What is the strongest case that this is not actually a problem?" A good tool will often produce a reasonable defense, and that defense tells you how confident the original finding deserves to be. Here is a short template you can adapt: - "Here is the genre and intended reader experience. Here is chapter by chapter what I meant to accomplish." - "For each scene, state: what the viewpoint character wants, what opposes them, and what changes by the end. Cite the passage that shows the change." - "List any scene where you cannot identify a clear change, and say why." - "Chart the dominant reader emotion per scene on a one to five intensity scale." - "Identify any region of three or more consecutive scenes with flat emotion and low change." That is a diagnostic, not a critique. It tells you where to look. Which tool you run it through is a separate question; Lacunai's catalog of AI models lets you compare how different ones read the same manuscript, and that comparison is itself useful. When two independent readers flag the same scene, take it seriously. When only one does, check it and decide. ### Working at manuscript scale One practical note on length. Eighty thousand words is a lot of text, and not every tool can hold all of it in view at once. If you are working in chunks, keep a running document of the findings so that each new pass has context from the previous ones. Otherwise the tool will analyze chapter nine without knowing what chapter two set up, and you will get false alarms about "missing" setup that actually exists earlier. If you are comfortable with it, a developer-minded writer can automate the bookkeeping through an API, feeding scenes in sequence and accumulating structured output. Lacunai's developer API is one way to build that pipeline. You do not need to. Many writers do the whole thing by hand with a spreadsheet and patience. But if you are revising several books a year, the automation pays for itself quickly. ## The Mistakes Writers Make With Developmental Editing Tools Having watched a number of people try this, I can name the failure modes. They are predictable, and none of them is about the technology. ### Accepting every note The tool says scene twelve is low-velocity. You cut scene twelve. It turns out scene twelve contained the only moment of genuine tenderness in the book, which is why the ending later hit nothing at all. A flagged scene is a question, not an execution order. Always ask what the scene is doing before you decide it is doing nothing. ### Rejecting every note The opposite error, and the more tempting one. The tool says your middle drags. You decide it is wrong, because you love the middle. Perhaps it is wrong. But if you have a pattern of dismissing every finding, you have not built a diagnostic process. You have built an expensive way to feel confirmed. A useful rule: if two separate instruments, or the tool and a human reader, flag the same region, you owe it a genuine look. ### Using it to avoid the hard decision Sometimes a writer knows, somewhere underneath, that the book's premise is the problem. The structure is fine, the scenes function, the pacing is acceptable, and the whole thing is still not working. No diagnostic will say that for you, because it is a judgment about whether the idea is strong enough, and that is a human call. Do not mistake a clean scene audit for a sound book. ### Skipping the human altogether I will say this again because it is the most useful thing in this essay. The point of the AI pass is to make the human edit better, not to replace it. A human developmental editor brings taste, market knowledge, and the ability to read what a book is trying to be rather than what it currently is. You are not buying less of that. You are buying a more refined version of it, because you arrive having already fixed the findable problems, which means the editor's hours go to the things only an editor can see. ### Revising in the wrong order Fixing the line-level prose before the structure is settled. The tool will happily polish your sentences if asked. Do not ask yet. Structure first. Always. ## What Happens When You Use This Before You Hire an Editor Let us come back to Elena, because her story has a useful ending. After her four passes, she had cut about nine thousand words, rewritten six scenes, restructured the antagonist's presence so the stakes were visible by chapter three, and fixed the sag in the middle by merging the two travel sequences and giving the second half of the book a deadline. The book was tighter and the middle moved. Then she hired a developmental editor. A human, and a good one. What changed was the nature of the conversation. The editorial letter did not spend three pages on scenes that did nothing, because those were already gone. It did not need to explain that the villain's plan was unclear, because Elena had fixed that. Instead it spent its length on subtler matters: whether Mara's decision at the two-thirds mark was fully motivated, whether the secondary romance was earning its page count, whether the ending's emotional register matched the book's opening promise. Those are the questions worth paying a skilled person for. She told me afterward that she felt she had paid for the interesting part of the edit rather than the obvious part. That is the right way to think about it. The AI pass is not a cheaper version of the same service. It is a different service that makes the expensive one more valuable. If budget is your constraint, and for most writers it is, there is no shame in doing the diagnostic work first and spreading your human spending across fewer, better-targeted interventions. Lacunai's pricing page lays out what access costs, and it is a rounding error next to a single round of a professional edit. Use the cheap tool to do the cheap thing well. Save the money for the thing only a person can do. ## A Few Honest Words on Craft and Trust I started by saying I wanted you not to throw things. Let me finish by saying what I actually want. I want you to win. I want your book to reach the reader you wrote it for, in the best shape you can get it into, without your having spent a year's savings to learn that chapter eight needed cutting. I think the tools that exist now can help with that, but only if you use them as instruments rather than oracles, and only if you keep your own judgment in the driver's seat. The writer who gets the most out of this process is not the one who trusts the model most. It is the one who asks the sharpest questions, checks every answer against the page, and is willing to be wrong about their own book. That last quality is the one no software can supply. It is also, not by coincidence, the quality that makes someone good at revision. So here is where to begin, and it is small. Take one chapter. Not the whole book. One chapter, the one you are least sure of. Write down what changes in each scene. Ask a model to do the same, and compare the two lists. Where they differ, read the passage again. That is the entire practice, scaled down to an afternoon. Do it once and you will understand the method. Do it across the manuscript and you will have the diagnostic that used to cost thousands and take weeks. Pour yourself something. Open the draft. Start with the chapter that worries you most. It usually is the one. The developmental edit nobody warns you about is coming either way. The only choice is whether you meet it first, on your own terms, with your own instruments, or whether it meets you in an agent's two-line rejection. Meet it first. --- Source: https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo Markdown alternate: https://lacunai.com/blog/the-developmental-edit-nobody-warns-you-about-and-how-ai-catches-what-yo.md