AI Writers Retreat  ·  AI & The Craft  ·  Issue No. 3
AI & the Craft — A Series for Writers

Talk to It Like
an Editor.

The quality of what AI gives you during revision has almost nothing to do with the AI. It has everything to do with what you hand it — and how you ask.

Prompting & Context Engineering for Revision
Part 3 of the Series

Imagine handing your manuscript to a reader — a sharp, attentive one — and saying only: "What do you think?" They'd do their best. They might notice something useful. But they'd be working without any idea of what you were trying to do, what you already know isn't working, what kind of feedback you can actually use right now, or what draft number this is. Their response would be shaped more by their own assumptions than by your actual needs.

This is exactly what most writers do when they bring a draft to an AI for revision help. They paste the text, type something brief, and hope for insight. Sometimes they get lucky. More often they get a polite, generic response that reads like a creative writing rubric.

The problem isn't the AI. The problem is the handoff.

This issue is about two things that fix that: prompting — the specific question you ask — and context engineering — everything you give the AI before you ask it. Together, they are the difference between a reader who says "this is interesting" and an editor who says "the tension breaks on page four because of this specific sentence."

You wouldn't hand a manuscript to a new reader and say "what do you think?" You'd brief them first. Do the same with AI.

The Prompt Is the Question. Context Is Everything Else.

Writers often treat these as the same thing. They're not. Your prompt is the specific ask — the question you need answered, the task you need done. Your context is the full picture you hand the AI so it can answer well: your draft, your intentions, your constraints, your reader, your concerns.

Think of it this way. If the prompt is the question you ask your editor, context is the conversation you have before you ask it. A great editor's note doesn't arrive in a vacuum — it arrives after the writer has said: here's what I was going for, here's where I think it's failing, here's what I've already tried.

AI responds to exactly this kind of briefing. Not because it "understands" your intention in any human sense — remember from Issue 2, it is predicting what comes next — but because rich, specific context narrows the probability landscape dramatically. It shifts the model away from generic literary feedback and toward observations that are actually relevant to the piece in your hands.

[Paste of draft] "What do you think of this scene?"
What you get

"This is a compelling scene with strong atmospheric details. The dialogue feels authentic and the pacing builds tension effectively. You might consider deepening the emotional stakes for the protagonist and ensuring the scene's purpose within the larger narrative arc is clear to the reader..."

Generic. Could apply to almost any scene ever written.

[Paste of draft] "This is the third draft of a scene where two estranged sisters meet at their father's house after his death. My intention is that the reader feels the affection underneath the argument — not stated, just present. I think I'm losing it somewhere in the middle. Don't give me general feedback. Read only for that one thing: where does the affection stop being legible, and why?"
What you get

"The affection is clear in the first exchange — particularly the detail about the younger sister using her sister's childhood nickname without seeming to notice. It starts to disappear around the line 'You always do this.' That phrase is familiar enough to signal intimacy, but the stage direction immediately after ('she turned away') reads as final rather than habitual. The body language is shutting the door instead of leaving it ajar..."

Specific. Earned. Actually useful in the revision.

The draft in both examples is identical. The AI is identical. The only difference is what the writer handed over alongside the text.

The Four Layers of Useful Context

Context engineering sounds technical. It isn't. It's just the practice of giving the AI the briefing your best reader would need. There are four layers worth building into the habit — click any of them to see what this looks like in practice.

1
The Draft
What you're working on and where it lives in the larger piece
Don't just paste the excerpt. Tell the AI what it's looking at. Is this the opening? A mid-story turn? The final scene? Chapter three of a novel or a standalone flash piece? The same paragraph reads differently depending on where it sits. A sentence that's too on-the-nose as an ending might be exactly right as an anchor in the middle.
"This is the opening scene of a 4,000-word personal essay. The full piece is about my mother's dementia. This scene takes place before the diagnosis — she is still herself, mostly. I'm including just this scene for now."
+
2
Your Intention
What you were trying to do — not what happened on the page
This is the most important layer and the one writers skip most often. Your intention is not a summary of the draft — it is the thing the draft is trying to become. What effect were you after? What should the reader feel, understand, or carry away? What is the central tension you're trying to hold? State this explicitly. The gap between your intention and the AI's reading of the draft is often the most useful piece of feedback you'll get.
"My intention is that the reader feels the mother's vitality very specifically — not as nostalgia, but as a present-tense fact. I want her to be funny and difficult at the same time. I don't want the scene to feel like a memorial."
+
3
Your Concern
The specific thing you already suspect isn't working
Every writer comes to a revision session with a nagging suspicion about their own draft. Name it. "I think the pacing collapses in the third paragraph." "I'm worried the narrator is too sympathetic too early." "The last line feels like I'm reaching." Naming your concern does two things: it directs the AI's attention precisely where you need it, and it gives you permission to ask a sharper question than "what do you think?"
"My concern is that the scene is too warm too quickly. She hasn't earned the reader's affection yet. I might be doing the sentimental work for them instead of letting the details do it."
+
4
The Scope
What kind of feedback you need right now — and what you don't
This is your most powerful tool for avoiding the generic. Narrow the scope of the feedback to match where you are in the revision. A first-pass structural read and a line-level copyedit are completely different tasks — don't ask for both at once, and don't leave it to the AI to decide which you need. Scope also means telling the AI what to leave alone. "Don't comment on the dialogue — I'm happy with it." "Ignore the grammar; this draft is still loose." These exclusions are as useful as the instructions.
"Read only for structure and pacing. I'm not ready for line edits. Tell me if the scene has a clear turn, and where you feel the energy drop. Don't comment on the dialogue or the ending — I know those need work."
+
You don't need all four layers every time. Even one — your intention, your concern — transforms the quality of feedback you get. Start there.

Asking the Right Question

Once you've built your context, the prompt itself — the actual question — should be as precise as a surgical instrument. The most useful revision prompts share a quality that good workshop questions also share: they are specific enough to be answerable and open enough to be honest.

"Is this good?" is not answerable. "Does the tension in this scene have a clear release, and if so, where does it happen?" is. "Make this better" is not a question — it's a surrender. "Read the last two paragraphs and tell me what the narrator seems to want, based only on what's on the page — not what I've told you about the piece" is an invitation to genuine reading.

Use the builder below to see how your context and your question work together.

Revision Prompt Builder
Your Revision Prompt

One Thing AI Can't Replace in This Process

All of this — the context, the precise question, the narrowed scope — is in service of getting feedback you can actually use. But there is a step that no prompt engineering can do for you, and it is the most important one: deciding what the work is actually trying to be.

The AI can tell you where your intention isn't landing on the page. It cannot tell you whether your intention is the right one. It can identify where the emotional register shifts. It cannot tell you whether that shift is a flaw or the whole point. It can read your draft as a resistant reader. It cannot tell you whether that resistance is something to overcome or something to honor.

Those judgments are yours. They are the reason the work exists. Context engineering and precise prompting give you better raw material for revision — clearer observations, more specific problems, more targeted questions. But the revision itself, the real one, the one where you decide what the story is about and fight for it sentence by sentence, remains entirely human work.

Which is, of course, why you're here.

AI can tell you where the intention breaks down. It cannot tell you what the intention should be.

END OF ISSUE NO. 3 — AI & THE CRAFT