The 30-Point Novel Revision Checklist: Structure, Character, Scene, Prose and Continuity

The fix is passes. Each pass looks for one class of problem, from largest to smallest, and ignores the rest. This checklist is organised that way: five passes, thirty checkpoints, in the order that avoids wasted work. It is the list referenced in our AI novel workflow, and it works the same whether you drafted alone or with a model.

Before you start: the read-through

Let the draft sit for at least two weeks. Then read it in as few sittings as you can, on a different device or in a different format than you wrote it in (printed, on a phone, as an ebook), and take notes only. No fixes. The notes you want:

• Where you were bored.
• Where you were confused.
• Where you did not believe a character.
• Which scenes you skimmed.
• What you remembered the next morning.

That last one matters. What you remember is what the book is about. What you do not remember is a candidate for cutting.

Pass 1: Structure (points 1–7)

Read the notes, not the manuscript. Make a scene list: one line per scene, with POV, location, what changes. Then check:

1. Does the book start where the story starts? Most first drafts begin one to three chapters too early. Find the first scene where the protagonist makes a choice that has consequences. Can the book begin there?

2. Is the inciting event before the 15 percent mark? If the reader does not know what the book is about by then, they are reading on trust, and trust runs out.

3. Does the midpoint change the game? At roughly the middle, something should shift the protagonist from reacting to acting, or reveal that the goal is not what they thought. If the middle is more of the same, the second half will drag.

4. Does every act end with the situation worse? Act one ends with commitment; act two ends with the lowest point. A second-act ending where things are going fine is the most common structural fault.

5. Is the climax the protagonist's choice? Not luck, not an ally, not the antagonist's mistake. The protagonist decides something, and the decision costs.

6. Does the ending answer the question the opening asked? Go back to the first chapter. What did it promise? The last chapter should pay for it, even if the payment is a refusal.

7. Are there scenes with no change? In the scene list, mark every scene where "what changes" is blank or is only "they travel" or "they talk about the plan." Each one is either cut or given a turn.

A tension map helps here: score each scene 1 to 5 for tension and plot the scores. Flat stretches longer than three scenes are structural, not local. This is a job AI does well; paste the scene list and ask for a score per scene with one line of justification, then argue with it.

Pass 2: Character (points 8–13)

For each major character, write their want, their wound, and how they are different at the end. Then check:

8. Does the protagonist want something specific? "To be happy" is not a want. "To get the deed back before the auction" is. If you cannot state it in one line, the reader cannot feel it.

9. Does the protagonist drive the plot? Count the scenes where the protagonist makes a choice versus scenes where things happen to them. Below about half, they are a passenger.

10. Does the antagonist have a coherent plan? Write it as a sequence of steps from their point of view. If the plan only works because the protagonist is slow, or changes to suit each scene, it needs rebuilding.

11. Does each major character change, or clearly refuse to? An arc is a change or a tested refusal. A character who ends exactly as they began, without that being the point, is furniture.

12. Does each secondary character have a function and a want? Function: what they do for the plot. Want: what they would be doing if the plot were not happening. Characters with a function but no want read as tools.

13. Do any two characters do the same job? Two mentors, two comic sidekicks, two rivals. Merge them. The merged character is always stronger.

Pass 3: Scene (points 14–20)

Now read the manuscript, scene by scene, and check each one against these:

14. Whose scene is it? One POV, and the POV character should be the one with the most at stake or the most to hide.

15. Does the scene start late and end early? Cut the arrival and the departure. Start with the problem already in the room; end on the turn, not on the reaction to the turn.

16. Is there a goal, an obstacle and a turn? The character enters wanting something; something prevents it; the scene ends with the situation changed. If any piece is missing, the scene is summary in disguise.

17. Is the setting doing work? A scene in a generic room could be anywhere. A scene in a specific place with specific pressures (a kitchen during service, a car in traffic, a church during a funeral) has built-in tension.

18. Does the dialogue have subtext? Characters who say exactly what they mean are not people. Check the dialogue guide for the tests.

19. Is exposition attached to conflict? Any paragraph of explanation should be there because a character needs the information now and someone does not want to give it.

20. Does the scene end on a reason to continue? A question, a decision, a reversal or a changed relationship. Not a character going to sleep.

Pass 4: Continuity (points 21–25)

This pass is mechanical and tedious, which is why it gets skipped and why readers find the errors. Keep the story bible open.

21. Physical facts. Eye colour, scars, height, handedness, the name of the dog. Search the manuscript for each character's name and read every physical description together.

22. Timeline. Build a dated list of every event, including offstage ones. Check travel times, seasons, pregnancies, healing wounds, the day of the week. A common error is a "three days later" that makes a Tuesday into a Sunday and a later scene at a bank impossible.

23. Knowledge. Who knows what, when. The most common continuity error in mysteries and thrillers is a character acting on information they do not yet have.

24. Names and terms. Every place, faction, object and term spelled and capitalised the same way throughout. Make a list and search for variants.

25. Rules. Magic, technology, law and money behave the same way in chapter 30 as in chapter 3. If a rule was broken for convenience, either fix the scene or establish the exception earlier.

This is the pass where AI is most useful and most under-used. Paste a chapter with the relevant bible entries and ask for contradictions; it is a local matching task that models perform well. Our guide to keeping AI consistent with your characters has the prompt.

Pass 5: Prose (points 26–30)

Only now, when scenes will not be cut, do you edit sentences.

26. Told emotion. Search for "felt," "realised," "knew," "wondered," and emotion nouns (anger, fear, relief). Each is a candidate for replacement with behaviour. See show, don't tell.

27. Repeated words and tics. Every writer has five words they overuse. Find yours with a frequency count and cut half the instances. Also check for repeated sentence openings ("She," "He," "The") in consecutive sentences.

28. Rhythm. Read a page aloud. If every sentence is the same length, vary them. If every paragraph is the same length, vary those too; uniform paragraph rhythm is one of the tells of unedited AI prose, and a common tell of tired human prose.

29. Filter words. "She saw," "he heard," "she noticed" put a layer between the reader and the experience. Usually the sentence is stronger without them.

30. First and last lines. Of the book, of each chapter, of each scene. These are the lines readers remember and the lines that decide whether they continue. Rewrite each one on its own.

If you drafted with AI, add a sixth mini-pass for its specific fingerprints. The AI prose editing guide has the list; the short version is that models default to certain constructions ("a testament to," "couldn't help but," the reflective final sentence that summarises the paragraph) that are individually fine and collectively unmistakable.

The checklist in one place

When to stop

There is no draft that cannot be improved, which means revision has to end by decision rather than by completion. A reasonable stopping rule: when a full pass produces only changes you could argue either way, the manuscript is ready for outside readers. After their feedback, one more structural look and one more prose pass, and then it goes out. Writers who revise past that point are usually avoiding the next book.

Try it in SparkLore

SparkLore's revision tools run the mechanical passes for you: a continuity check that compares a chapter against the bible entries for every character and rule it mentions, a tension score per scene from the outline, and a consistency report across chapters. The structural and character decisions remain yours, which is how it should be. If you want to try the continuity pass on a single chapter before creating a project, the free AI story continuation tool accepts an instruction line; paste a chapter with a few bible facts and ask it to list contradictions rather than continue. Sign up to run the checks across a whole manuscript.