Ask AI for feedback on an old draft without losing your intent
Before asking AI what to change in an old draft, decide why you reopened it. You may want to revive the whole piece, carry one scene into a new project, or simply understand what you once tried. Preserve the original, choose a small excerpt, and ask for observations tied to its sentences. Then test one change you can reverse. This keeps an AI response from turning a personal return to old work into an automatic full rewrite.
Read the old draft before asking for feedback
Start by making a copy of the old file and leaving the original untouched. Read a few pages without editing. In a separate note, write three short answers: What did this piece try to do? Which image, question, or character choice still interests me? What no longer fits the writer I am now? An old draft is evidence of an earlier creative decision, not an assignment to preserve every sentence.
Choose an outcome before opening an AI tool: revive this piece, borrow one element for a new piece, or archive it with a useful note. This avoids treating every AI suggestion as a demand for a full rewrite. The University of North Carolina Writing Center distinguishes substantive revision of purpose and organization from line editing; for an old draft, that order is especially helpful because polishing an opening makes little sense before deciding whether the piece still has a job. See [UNC’s revision guide](https://writingcenter.unc.edu/tips-and-tools/revising-drafts/).
Give the AI a narrow review job
If you want feedback, send only the excerpt needed for one decision. An opening page may help test whether the reader can understand the situation; a middle scene may help test whether a character’s choice is visible. A whole manuscript invites broad, repetitive advice and can hide the particular thing you want to learn. Check the service’s current data controls before uploading unpublished work, and remove details about other people that the task does not require.
Tell the model the historical context without telling it what conclusion to reach. For example: “This is an old fiction draft. I am deciding whether to revive its central character or borrow one scene for a new piece. Describe what the character visibly wants in this excerpt, quote the sentence that suggests it, and list what remains unclear. Do not rewrite the prose.” The exact task and bounded input follow the general prompting principle in [OpenAI’s prompt-engineering guide](https://developers.openai.com/api/docs/guides/prompt-engineering): specify the work to be done and provide relevant context. The guide does not establish that the model’s literary judgment is correct.
Sort feedback by the decision it changes
Read the response against the actual excerpt, not against a vague impression that the AI sounds confident. Mark each comment as supported, uncertain, or misplaced. A supported comment names a passage and a reader-facing effect you can inspect. An uncertain comment may be worth testing but needs your own judgment. A misplaced comment assumes a different genre, audience, or goal from the one you chose.
Imagine an old story in which a character spends a page repairing a broken model boat. An AI response might say, “The repair delays the plot.” That may be true if the scene’s job is to move quickly toward a departure. It may be false if the careful repair reveals a choice the character has avoided. Check what the scene does in the whole piece before cutting it. The same sentence can be excess detail in one project and the point of the scene in another.
Run one reversible experiment
Choose one supported or testable observation. Make a copy of the relevant page and try a single change: move the character’s decision earlier, cut one explanation, or restore a detail that clarifies the scene. Compare the old and new versions side by side. Do not ask the model to rewrite the entire draft before you have seen whether that local change helps.
Read both versions aloud. UNC’s [reading-aloud guide](https://writingcenter.unc.edu/tips-and-tools/reading-aloud/) recommends attending closely to the words on the page and noticing unintended effects. For an old draft, ask two questions after the read: Is the scene clearer to a reader? Does it still contain the element that made me return to it? A version can be more fluent and less worth writing; a rough line can still point toward a useful revision.
When the right answer is to borrow or archive
If the premise no longer attracts you, do not treat an AI’s ability to propose fixes as a reason to rebuild the whole piece. Save a brief archive note: “I still like the workshop setting and the dialogue on page four; the central plot is not one I want to pursue.” That makes the old draft findable without leaving an open obligation.
Borrowing one part is also a real creative decision. Copy the chosen passage or idea into a new project note, record where it came from, and let the new piece establish its own purpose. You do not need a verdict that the old draft was good or bad. You need a clear next action that belongs to the writer you are now.