Outlining a long novel with AI: the eight things to plan before you draft

Every novel-outlining guide on the internet hands you the same three-act diagram and tells you to fill in the boxes. Act one: setup. Act two: rising action. Act three: resolution. You stare at it. You write down "Maren learns the truth" in the middle box. You feel slightly worse than before you started. The diagram is not wrong. It is just not enough for a 100,000-word book. A long novel needs more than three boxes nailed down before you draft. The pages it takes to fail without a plan are too many to redo on instinct.

Here are the eight things a long book actually needs planned before chapter one, the order they tend to fall into place, and what AI can usefully do at each step. If you find yourself remembering the acronym halfway through, that is by design.

P: the promise

Before anything else, decide what kind of book you are writing.

A promise is the compact you make with the reader on page one. Genre. Tone. Pace. What kind of payoff they are owed. A literary character study and a courtroom thriller can share an identical premise and be entirely different books because the promise is different. Readers feel the breach when a book switches register on them in chapter eight, even if they cannot articulate what went wrong.

What to nail down: the genre, the tone (dry, lyrical, propulsive, tender), the rough comp titles, the reading speed you are pacing for, and what kind of ending you are honour-bound to deliver. Tragic. Triumphant. Open. Quiet.

AI is genuinely useful here. Hand it your premise and three comp titles and ask what promise that combination makes to the reader. The answer is often clearer than the one you arrive at alone.

E: the engine

The structural skeleton. Whatever beat plan or scaffold you trust to keep the story moving.

There are many engines. Save the Cat. Story Genius. The seven-point story structure. The hero's journey. A bespoke one you have built over five novels. Pick one. The point is not which one. The point is that without an engine, a long novel sags in the middle and ends three chapters too late, every time.

Map your premise onto the beats. You do not need every beat filled in. You need the load-bearing ones: the inciting incident, the midpoint reversal, the dark night, the climax, the resolution. The bones.

AI is useful for beat-mapping if you are honest about it. Give it your premise and a chosen structure, and ask where each beat lands. Treat the output as a draft, not the answer. The engine is your scaffolding. Live with the load-bearing beats for a few days before you commit.

N: the nexus

What the book is about.

Not the plot. The plot is what happens. The nexus is what the book means. The theme. The argument the book is making about being alive. A novel without a nexus reads as competent and hollow. Readers cannot name what is missing but they do not recommend it.

The nexus is the hardest layer to articulate before you draft. You may not know what your book is about until act two. That is fine. Write a placeholder. "This book is about whether forgiveness is owed to people who refuse to remember." Update it when you learn more.

The nexus belongs in every drafting decision afterward. Scenes that do not touch the nexus should justify their existence in some other way. Most do not, and most are cut later.

AI is poor at finding your nexus from scratch. It will give you generic themes. It is much better at pressure-testing a nexus you already have. Tell it the theme. Ask which characters embody it, which oppose it, and where the book might be currently silent on it.

D: the dynamics

The scene plan.

This is where most outlining stops, and it should not. A scene plan is more than a list of what happens. A useful scene plan names, per scene, who is in it, what they want, what changes, what the reader learns, and whether the scene's emotional charge runs positive or negative. Across a sequence of scenes the charges should alternate. Two positive scenes in a row are usually one too many.

You do not need every scene before you start. You need the first ten. After that the plan can grow as you draft.

AI is genuinely helpful at scene-level planning if you have committed the layers above. With a promise, an engine, and a nexus in place, the model can generate a credible scene list. Without them, it generates filler.

R: the registry

The cast and the world.

Every named character, every named place, every named object that matters to the plot, every rule of the world the reader needs to understand. Captured as structured records, not as a Word document.

This is the layer that catches the consistency failures every long-novel writer hits. Eye colour drift. Renamed factions. Geography that contradicts itself. A registry is the queryable cast and world that the drafting tool reads from and writes back to.

For long fiction, the registry is the single highest-leverage layer in the planning stack. A book with a thin engine and a strong registry can still finish coherent. A book with a brilliant engine and no registry usually cannot. Pendraic's Story Index is built for exactly this.

AI is useful for seeding the registry from a premise and outline. It will draft character records, place entries, and rule definitions you can edit. The edit step is required. The model will overshoot and invent details you did not ask for. Cull aggressively.

A: the anchors

The voice.

What your prose sounds like. Sentence rhythm. Vocabulary range. The way you handle dialogue. The writers you echo. The phrases you would never write.

Most outlining systems skip this entirely. Voice is treated as something that emerges during drafting. That works on a short story. On a 100,000-word novel it produces tonal drift between chapter one and chapter forty-seven, and the drift is worse if AI assistance is involved.

A voice plan is the smallest set of constraints that, applied consistently, keeps your prose recognisably yours. A 1,500-word sample of your finished work. A list of authors you sound like. A list of phrases banned. A note on register.

This is what Anchors does inside Pendraic. The voice profile is captured once and applied across every drafting and editing call.

I: the iterations

The revision system.

Not "how many drafts will I do." That answer is always wrong. The actual question is: how will I track what changes across drafts, why, and which version of any given scene is current?

Most writers manage this with file naming. Chapter 3 v4 final REAL final.docx. By draft three this collapses into chaos. Scenes get reverted by accident. Old versions get shipped to beta readers. The revision history dies in a folder.

The fix is a system that stores every iteration of every scene with the change rationale, so you can always retrieve the version you regret deleting and always answer "why did I change this?" Iterations is the layer for this.

You do not need to plan the contents of revisions in advance. You need to commit to the system before you draft, because retrofitting it is painful.

C: the closure

The setup-and-payoff ledger.

Every plot thread you open, you owe the reader a closing on. Every gun on the wall. Every romance subplot. Every mystery. Every secret. Half the failures in long novels live here. A reader cannot always name what was unresolved, but they feel it as the book ending unsatisfyingly.

A closure ledger is a list of every open thread with its current status. Open. Closing. Closed. Abandoned (with a note on why). Queryable any time, especially in act three when you are deciding what to wrap up.

You start the ledger at planning and append to it as you draft, because most threads are not visible in the outline. They emerge in scenes. Naming them when they appear is the discipline that keeps the book honest.

The acronym

Promise, Engine, Nexus, Dynamics, Registry, Anchors, Iterations, Closure. PENDRAIC. The eight things a long book actually needs nailed down. The full method writes up each layer in more detail. The platform organises a project around it so you do not have to assemble it from eight separate tools.

A three-act diagram is a useful start. A long novel needs the other seven.

What to try first

If you are outlining a new novel right now, you do not need to do all eight perfectly. You need the first three roughly right (promise, engine, nexus), the registry seeded with the cast and a few key places, and the voice anchored. The rest you can build out as you draft, if the tool you are using supports it.

Sign in if you want to plan a long novel against a system that holds all eight layers in one place. The drafting calls downstream will be working with a fuller picture than they ever get from a chat window and a blank document.

Outlining is not preparation for the work. Outlining is the work, done early enough that the draft can be honest.