Padding Isn't Always the Problem
Every writer knows the feeling. You reread a scene and think "this drags," so you start cutting. You trim dialogue tags, tighten description, maybe kill a paragraph of backstory. And the scene still feels bloated. Nothing you cut actually fixes it, because the problem was never length. The problem is that the scene is trying to do two jobs at once, and neither job gets done cleanly. Think about the classic culprit: a scene where two characters walk through a market so the author can show off the worldbuilding, and somewhere in that same scene, they're also supposed to have their first real emotional moment together. The prose isn't wordy for no reason — it's wordy because it's serving two masters. Half the sentences are doing set-dressing work. The other half are doing relationship work. Neither gets the space it needs, so the worldbuilding feels rushed and the romance feels unearned, even though word-for-word, nothing is badly written. This is a structural problem, not a prose problem, which is exactly why revision passes aimed at line-level fixes don't solve it. If you've worked through The Five-Pass Revision Order for AI-Assisted Novels, you know structural issues need to get caught before you're polishing sentences — otherwise you're sanding a chair that still has three legs.
A scene that feels "off" without an obvious error is usually a scene carrying more than one job. Your gut catches the overload before your eye catches the cause.This matters especially in AI-assisted drafts, because AI models are eager to please and tend to stack purposes into a scene rather than trust that a scene can do one thing well. Ask for "a scene where they explore the ruins and their tension boils over" and you'll often get a scene that rushes both. The fix isn't better prompting up front — it's a diagnostic pass afterward that names exactly what a scene is trying to accomplish.
Prompting AI to List Every Structural Job a Scene Is Doing
The first move is to stop asking AI "is this scene good" and start asking it to inventory the scene's structural workload. Vague quality questions get vague quality answers. Inventory questions get you a list you can actually act on. Paste the scene in full and ask for a breakdown by job category, not by paragraph. You want to know what functions are present, not just what happens.
Here is a scene from my manuscript [paste scene]. Don't evaluate the prose quality. Instead, list every structural job this scene is doing, choosing from: plot advancement, character reveal, worldbuilding/setting establishment, foreshadowing, relationship/romantic shift, thematic statement, tonal reset, pacing beat (breather or tension raise). For each job you identify, quote the specific lines that are doing that work and rate how much of the scene's total word count is devoted to it (rough percentage).
What you get back is often a surprise. Writers usually think a scene is doing one job — "this is the scene where they find the map" — and the breakdown reveals it's quietly doing three: plot advancement (finding the map), worldbuilding (explaining how the map system works in this world), and a relationship beat (a small moment of trust between the two characters holding it). None of those are bad ideas individually. Together, in one scene, they're fighting for oxygen. This diagnostic works best when you run it scene-by-scene rather than on a whole chapter at once — you want granularity. If you're doing this across a full manuscript, it pairs well with the Three-Act Beat Sheet tool, since you can cross-reference which structural jobs your outline assigned to a scene versus what the drafted scene actually ended up carrying. Drift between the two is often exactly where bloat creeps in. It's also worth doing this pass before you touch dialogue or voice at all — jumping straight to line edits on an overloaded scene means you're polishing a job you might cut entirely.
The Single-Job Test: What Does This Scene Do Best?
Once you have the inventory, the next question is sharper: which job does the scene do best, and which jobs is it doing badly as an accidental byproduct? This is the test that actually tells you what to keep.
Based on the structural job breakdown you just gave me, tell me which single job this scene would accomplish best if it were the ONLY job the scene had to do. Then tell me which of the other jobs are being done poorly or superficially because they're competing for space. Be specific about what's lost — for example, if worldbuilding is crowding out the romantic beat, tell me exactly which emotional beats get rushed or skipped as a result.
This question forces AI to rank rather than just list, and ranking is where the real diagnosis happens. You'll often find that a scene's "best" job isn't the one you intended when you outlined it. Maybe you wrote the market scene thinking it was primarily a romance beat, but the AI breakdown shows the worldbuilding is actually the strongest, most specific writing in the scene, and the romance is the part running on autopilot — generic banter, a stolen glance, nothing that couldn't be lifted into any other scene in the book. That's diagnostic gold. It tells you the romance beat isn't earning its place here, not because romance is wrong for this moment in the story, but because this scene's other job is stealing its attention. If you're working in genre where beat placement is unforgiving — romance especially — this single-job test is worth running against your genre's expected structure. The Romance Beat Sheet: Where AI Drafts Usually Break guide covers where these specific beats tend to get crowded out in AI-generated drafts, and it's a useful companion check once you've isolated the competing jobs. The same applies to worldbuilding-heavy genres. If you're working in fantasy or sci-fi, it's common for exposition and magic-system explanation to quietly eat scenes that were supposed to be about something else entirely. If your single-job test keeps surfacing "worldbuilding wins, character work loses," it might be worth running your scenes against the constraints outlined in Fantasy Magic Systems: Constraints AI Will Respect — sometimes the fix isn't cutting worldbuilding, it's tightening the system's rules so you need fewer words to convey them.
Split, Cut, or Leave Alone — A Decision Framework
Not every overloaded scene needs surgery. Some genuinely benefit from doing two things — tension and worldbuilding often reinforce each other when done well. The question is whether the two jobs are complementary or competing. Here's the framework I actually use:
- Split it when both jobs are structurally necessary and both deserve full weight — for example, a plot revelation and a major relationship shift that both need room to land. Cramming them together shortchanges both; splitting into two scenes (even short ones) lets each hit properly.
- Cut the weaker job when one job is clearly filler dressed up as purpose — worldbuilding that could be conveyed in a single sentence elsewhere, or a foreshadowing beat that isn't actually load-bearing later in the book.
- Leave it alone when the two jobs are genuinely reinforcing each other — tension rising while a secret gets revealed, for instance, where the reveal causes the tension rather than sitting beside it.
The trap most writers fall into is assuming every overloaded scene needs to be split. Splitting has a cost — it adds a scene, which adds pacing weight, and if you split every double-job scene in your manuscript you'll end up with a book twice as long and half as tight. Ask AI to make the case for cutting before you default to splitting.
I've identified that this scene is doing two jobs: establishing how the blood-debt magic system works, and marking the moment where Kessa starts to trust Rook. Make the strongest possible case for CUTTING the worldbuilding job entirely from this scene and delivering that information elsewhere (or not at all). Then make the strongest possible case for SPLITTING this into two scenes so each job gets full space. Tell me which option better serves pacing if this scene sits in chapter 6 of a 24-chapter fantasy novel where the blood-debt system becomes critical in chapter 19.
Giving AI the chapter position and the payoff distance (chapter 6 setting up chapter 19) matters enormously here — a scene's job isn't just about what it does locally, it's about when the payoff arrives. This is the kind of context a story bible should already have loaded in, which is why this diagnostic works best when you're not re-explaining your plot from scratch every time. If you haven't set one up, How to Build a Story Bible That AI Models Actually Follow walks through making sure AI actually has that kind of long-range plot awareness rather than just reacting to the scene in isolation. If you're leaning toward cutting, don't just delete — ask AI to find the smallest possible carrier for that information elsewhere:
If I cut the blood-debt magic explanation from this scene, suggest three alternate, much smaller ways to seed that information earlier or later in the manuscript — a single line of dialogue, a physical detail, or a piece of internal thought from Kessa. Each option should cost no more than two sentences.
This keeps you from losing worldbuilding entirely — it just relocates it to somewhere it doesn't compete with anything else.
Rebuilding the Scene: Drafting the Split Version for Comparison
If the decision framework points toward splitting, don't just outline the split in your head — have AI actually draft both versions so you can compare them side by side. This is where the diagnostic pays off, because reading two real drafts tells you things an outline can't.
Take the original scene [paste it] and split it into two separate scenes. Scene A should focus entirely on establishing the blood-debt magic system through Kessa and Rook navigating the market — cut any dialogue or beats that were serving the trust/romance arc. Scene B should focus entirely on the trust moment between them — strip out magic-system exposition and let the emotional beat breathe, using a different, quieter setting if the market doesn't serve it (a walk back to the inn, a shared meal, whatever fits). Keep both scenes under 900 words each. After drafting both, tell me what got lost from the original scene that I might want to preserve.
That last instruction matters — asking what got lost. Splitting a scene often strands a good line or image that only worked because it was doing double duty (a piece of dialogue that was both witty banter and a worldbuilding aside, say). AI will surface those casualties if you ask directly, and you can decide whether to rescue them into one of the new scenes or let them go. Once you have both drafts, read them back to back against the original. Nine times out of ten, the split versions are individually tighter even though the combined word count is similar or slightly higher. That's fine — page count isn't the enemy here, unfocused attention is. If you're watching your manuscript's overall length for pacing or publishing reasons, the Chapter Word Count Planner is useful for checking whether a split adds an acceptable amount of length to the surrounding chapter rather than blowing out your pacing plan. This whole diagnostic — inventory, single-job test, decision framework, rebuild — is something you can run as a repeatable pass across an entire manuscript, not just on scenes that already bother you. In fact, running it on scenes that feel fine is often where you catch the sneakiest overload, because a scene doing two jobs adequately doesn't trigger your internal alarm the way a badly written scene does. It just sits there, quietly underperforming, until a beta reader tells you chapter 6 "felt like it was missing something" without being able to say what. If you're building this into a full revision workflow rather than doing it ad hoc, it fits naturally alongside the broader AI novel writing workflow — run it as its own structural pass before line editing, and definitely before you get to a beta reader workflow, since beta readers are far more useful giving feedback on a structurally sound draft than diagnosing overload themselves. It's also a useful check no matter which genre tool you're drafting in, whether that's a romance novel with AI, a mystery novel with AI, or something in LitRPG writing where system explanations are especially prone to crowding out character beats. The mechanics of the diagnostic don't change — you're always asking the same four questions, just pointed at different genre pressures. For a broader look at where this fits in the editing process generally, how to edit a book with AI covers the surrounding revision stages this diagnostic slots into.
Next time a scene bugs you but nothing you cut seems to fix it, don't reach for the delete key. Paste it into a fresh chat and run the inventory prompt first — list every structural job, quote the lines, get the percentages. That single question will usually tell you within thirty seconds whether you're looking at a padding problem or an overload problem, and those two problems need completely different fixes.