One large prompt rarely produces a usable long-form draft. It produces something evenly competent that you then have to argue with. Staging the work creates checkpoints where a wrong assumption can be caught before it propagates through nine thousand words, and it keeps the structural decisions where they belong: with you, early, in writing.
- Replace the single mega-prompt with five stages and a checkpoint between each.
- Write unknowns in the brief as open questions instead of letting them be filled in.
- Label every note as evidence, explanation, example, objection, or action before outlining.
- Build the outline from reader questions and objections, one full sentence per section.
- Draft the section you know best first, and write the introduction last.
Why the single-prompt draft disappoints
A request for a whole piece forces hundreds of decisions at once — scope, order, emphasis, evidence, register — with almost no information to make them from. The result defaults to the most common shape for that topic, which is why assisted drafts feel interchangeable.
There is a practical cost too. When the whole thing arrives at once, an error in the second section has already shaped the seventh. Fixing it means unpicking connective tissue that was written to support a claim you no longer want to make.
Stage one: a decision brief on one page
Name the reader, the job the piece does for them, the promise you are making, the evidence you hold, the boundaries you will not cross, and the action you want at the end. Half a page is enough if every line is a decision. Doing that first is most of what keeps a draft sounding like you.
Where you do not know something, write it as a question rather than filling it in. An unanswered question in a brief is cheap. The same gap discovered in a final draft has usually been papered over with a confident sentence nobody can source.
Every stage you skip becomes a problem you meet again later, larger and harder to unpick.
Stage two: sort the raw material by function
Go through your notes and label each one: evidence, explanation, example, objection, or action. This is faster than it sounds and it exposes two things immediately — the arguments you have three examples for, and the ones you have only an opinion for.
An assistant is useful here as a second pass. Ask it to group the notes by theme and to list what a reader would expect but cannot find in the pile. Treat both as prompts for your own sorting, not the sort itself.
Stage three: outline on the reader's questions
Instead of an introduction-body-conclusion shell, list the questions your reader will have in the order they will have them, plus the objections they will raise if you dodge one. Sequence those. That sequence is your outline, and it explains itself as it goes.
Write each section as one full sentence stating what the section must establish, with the supporting notes linked underneath. Ask for two alternative orderings if you are unsure, then choose one yourself and write down why the other was worse.
Stage four: draft one section at a time
Work section by section with only that section's notes attached. Constrained context produces more specific prose, and it makes review tractable: you can read four hundred words against five notes and see whether the claims are carried by the material.
Start with the section you understand best, not the first one. Momentum comes from writing something you actually know, and the early sections are usually easier to write once the middle of the argument exists on the page. Missing notes show up early that way.
Stage five: seams, then the opening
Sections drafted separately meet awkwardly. Read the joins in one pass and fix only the transitions: what the reader now knows, what they still need, and whether any section repeats a point already settled. Repetition across seams is the usual defect.
Write the introduction last. By then you know what the piece actually delivers, so the opening can promise that rather than the piece you intended to write three days ago. The same applies to the title, the summary, and any subheading you drafted from the brief.
Checkpoints that catch drift early
At the end of each stage, ask one question: does this still serve the reader named in the brief? Drift is rarely dramatic. It looks like a section that grew because it was interesting, or an audience that quietly shifted toward your colleagues.
Keep a short running list of claims that still need a source and questions you deferred. Carrying that list forward is what stops a staged workflow from turning into a tidy process that produces an unverified draft on schedule. Process is not the same thing as rigor.
From an unsorted pile to a sentence outline
Notes for the middle section, in the order they were captured: interviewed six growers, four said the co-op price floor was why they switched crops. February co-op minutes describe the floor as temporary; it is still in place. Marta says the floor keeps the mill open, not the growers. Soil report — probably belongs in the last section? The mill shut for three weeks in 2024 and nobody I spoke to blamed the floor. I keep wanting to argue the floor is harmful but the growers do not say that. Need to check who actually sets it.
Section 3 promise: the price floor is doing a different job than everyone assumes. One, open on the four growers who switched crops and said the floor decided it (grower interviews). Two, establish that February's minutes called the floor temporary and it was never lifted (co-op minutes; verify who sets it). Three, turn on Marta's claim that the floor exists to keep the mill running (single source, needs a second). Four, test that against the 2024 closure, which nobody attributed to the floor. Five, close by naming what I cannot yet argue. Parked: soil report, for the final section.
The outline is not prose, but every line is a decision: what the section claims, which note supports it, and what is still unverified. Drafting from this is a different task than drafting from the pile above it.
Common mistakes
| The mistake | Why it costs you, and what to do instead |
|---|---|
| Outlining with generic headings | "Background", "Key considerations", and "Conclusion" describe furniture, not argument. Write each heading as the claim the section proves, then you can tell at a glance whether the sequence holds together or merely covers the topic. |
| Drafting with the entire note set attached | Attaching everything invites the assistant to reference material that belongs three sections away, which creates repetition you will spend the seam pass removing. Give each section only the notes that section needs. |
| Writing the introduction first | An opening drafted before the argument exists promises the piece you planned, not the one you wrote. It then quietly constrains the draft to match the promise. Leave a placeholder and return last. |
| Letting speed hide unresolved questions | A staged process can produce a complete-looking draft while every open question is still open. Keep the deferred list visible so finishing the prose is never mistaken for finishing the work. |
Your checklist
- Write a one-page brief naming reader, promise, evidence, boundaries, and desired action.
- Turn each unknown into a written question rather than an assumption.
- Label every note by function and identify which claims lack support.
- Build a sentence outline from the reader's questions, with sources attached per line.
- Draft section by section, starting with the one you understand best.
- Fix the seams, then write the introduction and title last.
Key terms
- Decision brief
- A short document that fixes reader, promise, evidence, boundaries, and desired action before drafting begins, and records anything still unknown as an explicit question.
- Sentence outline
- An outline where each section is one complete sentence stating what that section must establish, with the supporting notes or sources listed beneath it.
- Seam pass
- A revision pass that reads only the joins between separately drafted sections, fixing transitions and removing points that were settled earlier in the piece.
Working on a draft with AI right now?
Frequently asked questions
How long should the brief stage take?
+
Twenty to forty minutes for a substantial piece. If it takes longer, you are usually researching rather than deciding, which is a different stage. Stop when the reader, the promise, and the boundaries are written down and the unknowns are recorded as questions.
Should I ask for the whole outline at once?
+
Ask for two or three candidate structures and then assemble your own from the strongest sequence in each. The value is in seeing alternatives you would not have considered, not in adopting one wholesale. Write down why you rejected the runner-up.
What if a section refuses to come together?
+
A section that refuses to come together is usually an evidence problem wearing a drafting costume. Go back to the notes for that section and check whether the claim is supported. If it is not, either research the gap or narrow the claim until your material can carry it.
Can I compress the stages for shorter pieces?
+
Yes. For anything under about a thousand words, merge sorting into outlining and skip the separate seam pass. Keep the brief and keep drafting the strongest section first. Those two habits carry most of the benefit at almost no cost in time.
How do I stop the draft from sounding assembled?
+
Read it start to finish in one sitting with no editing, marking only where the voice changes or an argument restarts. Then revise those places by hand. Staged drafting produces seams; a single continuous read is what reveals them. Nothing else finds them reliably.
Where does research fit in this workflow?
+
Before sorting, and again after outlining, when the gaps are specific. Researching mid-draft is the most common way to lose an afternoon, because an open question in a half-written section always looks more urgent than it is. Note it and continue.