A practical AI writing workflow, from first draft to revision
Work in four steps. 1) Brief: state the audience, format, length, tone and the phrases to avoid. 2) Draft: ask for the full piece, not an outline, so structural problems show. 3) Critique: ask the model to list where the draft misses the brief before it rewrites anything. 4) Revise: fix the listed problems one at a time in the same thread, then verify every fact yourself.
Most bad AI writing comes from a bad first instruction. If the model does not know who is reading, how long the piece should be or what it must not say, it fills the gap with generic prose — and you spend longer editing than you would have spent writing.
The steps below are the loop itself, with the prompt to use at each stage. They work in any capable chat tool; where a ChatOTL feature makes a step easier, it is noted.
Step 1 — Write the brief in one message
Before asking for a single sentence, state five things: who is reading, what format it is, roughly how long, what tone, and what must not appear. This one message removes most of the editing you would otherwise do. If you write the same kind of piece repeatedly, save the brief as a reusable preset instead of retyping it.
Step 2 — Ask for the full draft, not an outline
Outlines look fine and hide structural problems. A full draft shows you where the argument thins out. Attach your source material at this point — notes, transcripts, a spec — so the draft is built from your material rather than the model's training data.
Step 3 — Run a critique pass before any rewrite
- "List where this draft misses my brief. Do not rewrite yet."
- "Which paragraph is weakest, and why?"
- "Which claims here would need a citation?"
- Switching to a reasoning model for this step often catches problems a fast model glosses over.
Step 4 — Revise one problem at a time
Apply the critique notes you agree with, individually, in the same thread so context accumulates. A single "fix everything" prompt usually flattens the prose. If you want a materially different angle, branch the reply into a new thread — the original draft stays intact so you can compare instead of gambling.
Before you publish
- Verify every fact, figure, quote and name yourself — no AI tool can do this for you without web access.
- Read it aloud; that is where borrowed rhythm and filler phrasing become obvious.
- Check the claims you did not write. Anything you cannot source, cut.
Where it fits
Long-form
Brief once, then draft section by section in one thread.
Voice matching
Attach three samples and ask for a reusable style sheet.
Hard cuts
Ask for 40% shorter, then audit what disappeared.
Example prompts
- Audience: technical buyers. 800 words, plain English, no marketing superlatives. Draft the full piece.
- Critique your last draft against my brief. List the misses; do not rewrite yet.
- Rewrite paragraph three only, keeping my sentence rhythm from the attached sample.
Limitations
- No web browsing, so no step in this loop can research current facts, prices or sources.
- Model output can be confidently wrong; step 4 is not finished until you have verified it.
- Voice is imitated from samples you supply — the model does not know your voice on its own.
Frequently asked questions
What is the best order of steps for writing with AI?
Brief, draft, critique, revise. The critique step is the one most people skip, and it is the one that saves the most time: asking for a list of misses before any rewrite keeps you in control of the prose.
How do I stop AI writing from sounding generic?
Constrain it before it drafts: name the audience, the length, the tone and the phrases you never want to see, and attach two or three samples of your own writing to match rhythm against.
Should I ask for an outline first?
Usually not. Outlines look fine and hide structural problems. Asking for the full draft surfaces where the argument thins out, and the fix costs one more prompt.
What should I check before publishing?
Every fact, figure and quote, plus any claim you did not personally write. Read the piece aloud to catch filler phrasing, and cut anything you cannot source.
Run the loop on one real piece
Brief it properly once and see how much of the editing disappears.
Try this workflow