How to Ask AI for Story Critique You Can Actually Revise From

Define the revision question before submitting the story

“Critique my story” leaves the criteria open. Instead, name the effect you want and ask where the supplied text supports or interrupts it.

The University of North Carolina Writing Center’s guide to getting feedback (https://writingcenter.unc.edu/tips-and-tools/getting-feedback/) recommends explaining your stage in the writing process and requesting the kind of response you need. Although its examples concern academic writing, that principle transfers usefully to fiction: tell the reader what you are currently trying to solve.

Choose a question narrow enough to answer from the passage:

Avoid building your conclusion into the question. “Why is this scene boring?” presumes a defect. “Where does the scene’s situation change, and what happens between those changes?” asks for information you can inspect.

Add a short statement of intent: “This is a quiet scene about completing a difficult craft project. I want steady progress, with one meaningful setback.” That gives you a standard for judging suggestions. A recommendation to add another setback may be reasonable in some stories but unnecessary for this one.

Supply enough context to make the answer meaningful

Include the full unit needed to answer your question. For a local clarity issue, that may be several paragraphs. For a scene’s turning point, include the setup, the choice, and its immediate result. A passage that ends before the choice cannot establish whether the choice is adequately prepared.

Label three things separately:

That separation matters when checking the response. If you explain a character’s goal in your notes, you still need to know whether it appears clearly in the prose. Ask the AI to identify whether each finding comes from the excerpt or your background notes.

Anthropic’s official prompting guidance (https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) recommends explicit instructions, relevant context, and clear output constraints for Claude. Apply those principles here by naming the task and the response format. Their usefulness in this workflow does not establish that every AI system will follow them consistently.

For a longer manuscript, keep each critique’s scope explicit. A scene-level response can identify a missing explanation within that scene; it cannot establish that the explanation is absent from chapters you did not provide.

Separate observation, interpretation, and suggestion

Ask for an initial evidence pass before requesting edits. Use three distinct categories:

For example, “The scene starts too slowly” combines an evaluation with an unstated standard. A more usable response would identify what happens before the character’s goal becomes explicit, explain why that placement might matter to your stated purpose, and leave the revision choice open.

Request a plausible alternative reading, too. A delay may create useful anticipation. A repeated object may establish a pattern. You need to know whether a proposed change would remove something intentional.

Include one evidence-backed strength to preserve. “The description is vivid” gives you little guidance. “The recurring scrape of the chair marks each interruption in the conversation” identifies a specific feature you can protect during revision.

Check the evidence before accepting the finding

AI output can contain claims that are incorrect or inconsistent with the supplied material. Anthropic’s official guidance on reducing hallucinations (https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) recommends allowing uncertainty, grounding claims in quotations, and checking claims against sources. It also states that these methods do not eliminate errors.

For story critique, make the draft the source. Before acting on a finding, check four things:

Claims of absence need special care. One quotation cannot prove that a goal is never stated. If the AI says a detail is missing, search the supplied passage yourself. Have it distinguish “not present in this excerpt” from “not present in the story.”

When evidence is wrong, repair the critique before editing the draft. A useful follow-up is: “The quotation you supplied does not appear in paragraph 6. Recheck this finding against the excerpt. Provide exact support or withdraw it.” If support remains missing, set the finding aside.

An accurate quotation can still support an unhelpful recommendation. A paragraph of description might delay an action while establishing the object that makes the ending work. Evidence makes a critique inspectable; your intended effect determines whether the proposed change is worth making.

Turn one finding into a reversible edit

The following original miniature example illustrates the process. It is not a measured result or an actual AI critique.

[1] Nia set the blue cup beside the kiln. Its handle had tilted during drying. [2] She checked the shelf labels, swept the bench, and counted the empty boxes. [3] The exhibition collection was at noon. She took up the trimming tool.

Suppose the revision question is: “Does the practical deadline become clear early enough to give the opening direction?”

Observation: The collection deadline first appears in paragraph 3, after the maintenance actions in paragraph 2.

Interpretation: Until paragraph 3, those actions may read as routine preparation because the passage has not yet named the deadline.

Alternative reading: The delayed deadline may deliberately change how the reader understands the preceding actions.

Possible edit: Move “The exhibition collection was at noon” to the beginning of paragraph 2.

This edit tests the placement of one piece of information. Save the original, move the sentence in a duplicate, and reread both versions. Ask whether the earlier deadline gives the routine actions useful direction and whether losing the later reveal makes the passage less effective.

Reversibility includes understanding what you changed. Moving one sentence lets you compare one clear variable. Moving the sentence, deleting the bench detail, and rewriting the ending together makes it harder to identify which change produced the result you prefer.

Use this decision aid when choosing among findings:

This is a practical editorial decision aid, not a validated scoring system. Its purpose is to connect each change to a reason you can explain and revisit.

Use this AI story critique request template

Replace the bracketed fields before sending. Start with the evidence pass; request revision options only after checking its findings.

Please critique the supplied fiction excerpt to help me answer one revision question.

Revision question: [One specific question about this passage.]

Draft stage: [For example: scene revision, before sentence polishing.]

Intended effect: [What I want this passage to accomplish.]

Features to preserve: [For example: restrained narration, the final image, or the current event order.]

Prior context: [Relevant facts established earlier, or “none supplied.”]

EXCERPT START [Paste the excerpt with numbered paragraphs.] EXCERPT END

First pass: evidence and interpretation only.

- Observation: a checkable feature of the excerpt. - Evidence: an exact short quotation and paragraph number. - Interpretation: the possible effect on how the passage reads. - Alternative: another plausible reading, where applicable. - Limits: missing context or uncertainty that affects the finding.

After checking the first response, send this follow-up:

Focus only on finding [number], which I have checked against the draft.

Offer two revision options:

For each option, explain the intended benefit, the tradeoff, and what I should compare between the original and revised versions.

Describe the changes without rewriting the passage. Preserve [features]. Include keeping the original if that remains a reasonable choice.

Finish with a revision decision

Choose one supported finding that matters to your current question. Record the proposed change, its intended benefit, and a reason to undo it. For the kiln example: “Move the deadline earlier so the routine actions have context; revert if the delayed reveal creates a more effective turn.”

Then make the edit yourself and read beyond the changed sentence. Check the transition into it, the paragraph after it, and any later passage that relies on the original arrangement. If another reader is available, give them the same focused question without telling them which version you prefer.

Stop when you can name the next action: keep the passage, test a specific change, or gather missing context. A useful AI critique leaves you with a revision decision grounded in your own text.

Keep exploring this topic