6 min read

Writing Faster With AI: The Parts of the Job It Can Take

Drafting is a small fraction of writing a novel. Here is where the year really goes, and which of those hours a model can give back to you.

Ask a novelist why the book took four years and they will tell you about the writing. Then you look at the four years, honestly, and the writing is not where they went.

They went into rereading chapter seven to check what colour her coat was. Into the fortnight after a work trip, spent recovering the thread. Into the six weeks of knowing the middle was wrong and not knowing where. Into the reread of the whole manuscript, twice, to answer one question.

Almost none of that is writing. All of it is overhead, and overhead is the thing a machine can take.

This is a different argument from the usual one about AI and fiction, which is about whether a model should generate prose. (It shouldn't, and that argument is here.) This is about the other eighty-five percent of the job: the remembering, the searching, the checking, the re-orienting. That work is real, it is exhausting, and it is not where your talent lives.

Where the time actually goes

Take a novel that occupied two years of evenings. A rough but honest breakdown:

Activity	Share of the hours
Producing new sentences	small
Re-reading your own book to find something	large
Getting back up to speed after a gap	large
Structural diagnosis (what's wrong, and where)	large
Continuity checking	moderate
Mechanical cleanup	moderate

The exact numbers vary by writer, and I'm not going to invent precise ones. The shape is what matters, and the shape is consistent: the minority of your hours produce the book, and the majority of your hours service it.

Every hour in the second group is an hour of your attention spent on a task that requires no judgment, no taste, and no voice. Which is exactly the profile of work that should be automated, in any field.

The four accelerations

1. Never re-read to remember

The most expensive habit in novel writing is the lookup. You need to know whether the brother was mentioned in act one, so you open chapter three and start skimming, and forty minutes later you have your answer and you have also lost the scene you were writing.

The cost is not the forty minutes. It's the context switch: you left the drafting mind and entered the searching mind, and the drafting mind takes a long time to come back.

A model that has read your chapters answers this in four seconds, in a sentence, without you leaving the page. That is not a small convenience. It's the difference between a writing session that survives a question and one that doesn't.

2. Recover the thread in minutes, not weeks

The single biggest killer of novels written around a job is not lack of time. It's re-entry cost. You get ten days away. When you come back you don't know where you were, so you reread. Rereading turns into editing what you reread. Nothing new gets written, and then the next gap arrives.

A summary of where you left off — what changed in the last three chapters, whose viewpoint you were in, what question was open, what you told yourself to do next — collapses re-entry from a week to a coffee. It is a boring, mechanical, enormously valuable piece of work, and it is a summarising task, which is the thing language models are unambiguously good at.

3. Find the structural problem in a day instead of a season

You know the middle is wrong. Finding where it's wrong is normally a full reread with a notebook: several evenings, and you'll still miss things, because you cannot see the shape of a book from inside its sentences.

The fast version: have the chapters summarised, one line each, and read the list. The sag appears immediately as a run of chapters where nothing changes. The thread that dies in chapter nine is visible as an absence. Eight weeks of vague dread becomes an afternoon of specific problems.

The model is not fixing your structure. It's rendering it, at an altitude where your own judgment finally has something to work on.

4. Stop doing the mechanical pass by hand

Repeated words. The tic where everyone nods before speaking. Filter words. Name spellings that drifted. Your character's eyes changing colour in act three.

This is proofreading, it is deadening, and it degrades the longer you do it, because attention is finite and you spend yours on it at exactly the moment you need it for the last revision. Hand it over without a second thought. Nothing is lost, because nothing here required you.

The speed-ups that are traps

Being honest about the other side, because "faster" is easy to sell and easy to mistake.

Generating the draft. Yes, it produces text quickly. No, it does not produce a book faster, because the generated paragraph has to be revised into your voice, and revising someone else's competent prose into your own is slower than writing yours, not faster. You've traded a blank page for a wrong page, and the wrong page is harder to fix, because it's already fluent enough to seem finished.

Generating ideas for the plot. A model will happily give you ten possible next scenes. They will be the ten most probable next scenes, which is to say the ten a reader has already seen. Speed here buys you mediocrity at a discount.

Letting it tell you the chapter is good. Instant feedback that is worthless, because it has no taste and it wants to please you. Faster reassurance is not faster progress.

The pattern holds: anything that speeds up the servicing of the book is a real gain. Anything that speeds up the making of the book is a loan against quality, and the interest is charged in revision.

What "faster" actually means

Not more words per hour. That was never the bottleneck.

It means more of your hours spent on the part only you can do. A writer who gets four hours a week and spends two of them rereading to find her footing is a writer with two hours. Give her those two hours back and you have doubled her novel, without her typing any faster.

That is the whole of the speed argument, and it's why the interesting tools are pointed at the manuscript rather than at the blank page.

The habit that makes it work

Keep the machine on the reading side of your process and keep it there deliberately. Ask it what is in your book. Ask it what changed, what contradicts, what went quiet, what you said you would do next. Do not ask it what should happen, and do not ask it to say it for you.

Do that, and the year you were going to spend servicing the manuscript becomes the year you spend writing it. Which is, presumably, the reason you wanted the time.

Narrale is built on exactly this division: a Story Bible the AI populates by reading your chapters, continuity and plot tracking that flags contradictions across the whole book, editorial lenses that report without rewriting, and a drafting page that stays empty, quiet, and entirely yours.