When a draft sounds finished
The draft has a confident opening, a vivid room, dialogue and a final sentence that sounds important. Yet when you describe what changed, the answer is thin. Two people discussed a danger they already understood and agreed to investigate it later. The scene has the appearance of movement without consequences that make the movement matter.
Surface fluency can make a passage easier to accept than to inspect. This is not a reliable test of whether AI produced it. Human writers make the same mistakes, and generated prose can contain effective choices. The editorial task is to establish what this passage accomplishes in this story.
Inspect cause before atmosphere
Write a plain account of the action: someone wants something, encounters resistance, acts, and leaves with a changed situation. Not every literary scene follows that sequence, but the exercise exposes passages whose only development is a more emphatic description of their starting mood.
In an invented example, a researcher finds an altered laboratory log. One continuation has her announce that something is wrong and promise to return tomorrow. Another requires her to choose between preserving the log as evidence and using it to stop a procedure, revealing her unauthorized access. The second is not automatically better prose. It creates a conflict whose result can affect what happens next.
Ask which consequence your version needs. If she can preserve the evidence, stop the procedure and remain undetected at no cost, the scene may have removed its own pressure. Adding shadows or urgent dialogue will not restore the missing choice.
Check whether emotion is earned
Betrayal can be declared before the relationship has been established. Forgiveness can arrive because the conversation needs to end. A character can recognize a profound truth that never alters a subsequent decision.
Return to the evidence in the scene. What did the character expect? Which action defeats that expectation? What can they no longer do comfortably afterward? Sometimes the repair belongs in an earlier scene establishing the trust being broken. Sometimes it means removing an interpretation the event cannot support.
Reflection and summary can be powerful. Do not turn this into a requirement that every emotion appear as a gesture. Ask whether the chosen form gives the moment its necessary weight.
Read rhythm across a page
A short sentence can land with force. A page of equally forceful sentences makes every observation compete for the same emphasis. Repeated two-line paragraphs, mirrored exchanges and concluding declarations can flatten a scene even when each sentence seems acceptable. The solution is not a fixed quota of long sentences.
Read aloud and mark where the thought changes. Keep connected perceptions together when a character works through them. Break paragraphs when situation, attention or pressure changes. Let dialogue include evasion and unequal investment where appropriate, rather than giving every speaker a balanced answer.
Examine each image for its job. Does it locate the viewpoint, sharpen a perception or create a meaningful association? An ornate comparison that only restates that a room is sad may be decoration the scene does not need.
Evidence with boundaries
Doshi and Hauser's 2024 short-story experiment found that access to generated ideas could improve evaluators' ratings while increasing similarity among stories. The finding permits benefit and concern together. It does not demonstrate that fluent generated prose reliably produces strong novels. [1]
An assistant's praise is also not independent evaluation. Ask it to identify the strongest counterexample to its assessment, cite the responsible passage and explain what a proposed change would cost. Check the answer: a critique can be eloquent and still misunderstand a scene.
Revise with a stopping point
Make three passes through one scene. First, identify the change in the situation and repair missing causes. Second, check whether knowledge and emotional responses follow from what happened. Third, revise for clarity, rhythm and specificity. Retain the earlier version so improvements can be compared rather than remembered selectively.
Give a reader both versions without labeling one improved. Ask what they believed, where attention moved and what they wanted to know next. Their response may favor the original. Keep what serves the story, document unresolved problems and stop when further changes become preferences rather than repairs. Endless polishing can be another way to avoid deciding what the scene is for.
Sources & editorial note
Prepared with AI research and drafting assistance. PennedCanon is the editorial publisher and has a commercial interest in the workspace. Examples are original illustrations, not recorded model experiments. Practical exercises are editorial recommendations, not guarantees of quality, authorship or copyright.