AI Skills for Writers: Draft, Edit, and Publish with AI

Writing isn't one task. It's roughly six, and they have very different relationships with AI.

Figuring out what you think. Deciding what to include. Getting words down. Making the words good. Cutting. Adapting for a different audience or format.

Handing over the wrong one is what produces work that's fluent and empty. Handing over the right ones is what gets a week's work into two days without anyone noticing a difference — except that you had time to do the hard part properly.

The stages worth delegating

Adapting. One finished piece into five formats. Nearly free, entirely mechanical, and the place most writers lose their evenings.

Cutting to length. Take 1,200 words to 800. AI is good at this, and it's the stage writers are worst at, because everything feels necessary when you wrote it.

Critique. Paste your draft and ask what's weakest, what a hostile reader would attack, where you've asserted rather than shown. This is the highest-value use available to a writer and the least used, because it feels like admitting something. Nothing of yours gets overwritten — you're getting an editor's read in ten seconds.

Research scaffolding. What are the standard counterarguments here, what am I obviously missing, who disagrees with this position. Leads, not content.

The stages worth guarding

Figuring out what you think. If you outsource this, you end up assembling something rather than arguing something, and readers can tell — not because the prose is worse but because there's no position in it.

The sentences that carry the piece. The opening, the turn, the line you want quoted. These are a tiny fraction of the word count and most of the value.

The decision about what to leave out. AI is systematically inclusive. Asked to cover a topic, it covers the topic. A good piece is defined as much by omission, and omission requires a point of view.

Skill: brief it the way you'd brief a subcontractor

A one-line request produces a generic draft, and then you spend longer fixing tone than you'd have spent writing it. A real brief — audience, purpose, what they already know, what you want them to do, what to avoid, roughly how long — produces something you can work with.

The brief takes four minutes. It's the difference between AI saving you time and AI costing you time while feeling productive.

Skill: critique in loops, not rewrites

The pattern that works: draft, ask for critique, decide which criticisms you accept, revise yourself, repeat.

The pattern that doesn't: draft, ask for a rewrite, accept the rewrite. That converges on the model's register within two rounds, and you won't notice because each round improved something.

Bea — the drafts that were costing her time

Bea is a freelance content writer with six regular clients. She started using AI for first drafts and found her output roughly unchanged — the drafting got faster and the editing absorbed all of it.

She tracked where the editing time went for a fortnight. Almost all of it was tone: the drafts arrived in a register none of her clients used, and fixing register meant rewriting sentence by sentence, which is slower than writing.

She inverted it. She now writes the first draft fast and badly — voice intact, structure loose — and uses AI to tighten, cut, and stress-test. Same tools, opposite direction. Her throughput went up by about a third and the client edits went down.

Ollie — the critique he didn't want

Ollie writes long-form for a technology publication. He resisted AI editing for a year on the reasonable grounds that his judgement was the product.

What changed his mind was using it only for hostile reading: paste the draft, ask what a sceptical expert would object to. On a piece he considered finished, it flagged that his central claim rested on a single example presented as a pattern.

He disagreed with most of the other points. That one was right, and he'd read the piece six times without seeing it — because he knew the other examples existed. He just hadn't put them in.

He never asks it to rewrite anything. He asks it to attack, and argues back.

The one thing

Use it early for structure and late for pressure. Keep the middle — the part where you decide what you actually think and find the words for it — because that's the part readers came for.

If you're spending more time fixing tone than you saved on drafting, you've handed over the wrong stage.