Skip to content
Home / How to Give AI Feedback That Improves the Next Draft

Editing

How to Give AI Feedback That Improves the Next Draft

“Make it better” isn’t feedback — it’s a wish. Here’s how to point at the exact sentence, show the fix, and close the loop so each draft actually gets better.

A woman marking up a printed draft with a red pen, laptop beside the pages.
A writer marking up a printed draft with a red pen, laptop at hand — the feedback pass that turns a decent draft into a better one. Photograph: Pexels.

Here’s a sentence I’ve typed more times than I care to admit: “Make it better.”

And here’s what came back, every single time: a draft that was different, but not better. Slightly longer. Slightly smoother. Still wrong in exactly the ways that mattered.

That was my fault, not the tool’s. “Make it better” isn’t feedback — it’s a wish. The AI can’t see what I see, and it can’t fix what I won’t name. Once I started giving feedback the way I’d give it to a human writer — specific, shown rather than told, one thing at a time — the second drafts started coming back genuinely improved. Sometimes the third draft is the one I publish. That’s not a failure of the tool. That’s the process working.

1. Point at the sentence, not the feeling

Vague feedback produces vague revisions. “This section feels flat” tells the tool nothing it can act on, so it guesses — usually by adding adjectives, which is the opposite of fixing flatness.

Specific feedback names the place and the problem. “The third paragraph reads like a list of features, not a reason to care. The reader doesn’t know why any of this matters to them.” That’s something a writer — human or machine — can actually do something with.

My rule: never send feedback that doesn’t include a quote. Copy the exact sentence or paragraph that’s bothering you into the message, then say what’s wrong with it. It takes thirty seconds longer and it changes everything, because now the tool is working on the real problem instead of its guess about the problem.

2. Show the fix, don’t just describe it

Describing a tone is hard. Showing it is easy. When I want a draft to sound more direct, I don’t write “make it punchier” — I write one replacement sentence in the style I want, and say “write the rest of this section like this sentence.”

The tool is a pattern machine. Give it the pattern and it will follow it. Ask it to invent the pattern from an adjective and it will reach for the most generic version of that adjective — which is how “punchier” turns into three exclamation marks and a sentence starting with “Picture this.”

Same thing works for structure. Don’t say “reorganise this.” Say “open with the mistake, then the fix, then the example — like the second paragraph of the draft I edited last week.” If you have an example of the thing done right, paste it in. One good example beats a paragraph of instructions.

3. One change per message

This is the one I resist most, because it feels slow. I want to send a list: fix the intro, cut the repetition, change the tone, add an example, shorten the conclusion. So I do — and the next draft fixes two of the five, half-fixes one, and quietly makes the other two worse.

AI tools are bad at juggling a list of unrelated edits. Each instruction competes with the others, and the weakest ones get dropped or mangled. So now I send one change at a time, in order of importance. Fix the structure first — there’s no point polishing sentences in a section that might get cut. Then the tone. Then the details.

It feels slower. It isn’t. Five focused rounds get me a publishable draft; one kitchen-sink round gets me a draft I have to rewrite by hand, which is slower than all five rounds combined.

4. Tell it what to keep

Feedback is almost always a list of what’s wrong. That’s natural — wrong things are what catch your eye. But a revision pass that only hears about problems will “fix” things that weren’t broken.

I’ve watched this happen: a draft has one great line, buried in a weak paragraph. I complain about the paragraph. The next version rewrites the whole thing, and the great line is gone, replaced bysomething smoother and forgettable.

So now every feedback message includes a keep-list. “Keep the opening sentence exactly as it is. Keep the example about the missed deadline — it’s the best thing in the draft.” It takes one line, and it protects the parts that are already working from the revision process itself.

5. Give it the rule, not just the correction

Here’s the difference between feedback that fixes one draft and feedback that fixes every draft after it. “Change ‘utilise’ to ‘use’” fixes one word. “I never use Latinate words when a plain one exists — ‘use’ not ‘utilise’, ‘help’ not ‘facilitate’, ‘start’ not ‘commence’” fixes a habit.

When you catch a recurring problem, stop correcting instances and start stating rules. I keep a short list of mine — no filler openers, one idea per sentence, examples before abstractions — and I paste the relevant rule into the conversation whenever the draft breaks it. After a few rounds, the drafts start arriving with the rules already applied, because the conversation carries the memory of what I care about.

This is also how you stop feeling like you’re starting from scratch every time. The feedback accumulates into a working style guide, written in your words, built from your actual edits.

6. Feed your edits back in

The most powerful feedback you can give an AI tool isn’t a comment — it’s your edited version. When I finish editing a draft by hand, I paste my version back into the chat and say: “This is what I changed and why. The opening is shorter because the reader already knows the problem. The third paragraph is gone because it repeated the second.”

That does two things. First, it teaches the tool your actual taste, not your stated taste — what you kept, what you cut, what you rewrote. Second, it gives you a record of your own editing instincts, which makes you a better editor of the next draft.

I do this at the end of every piece now, even when the draft is already good enough to publish. It costs two minutes and it makes every future first draft a little closer to the one I’d write myself.

7. Check the next draft against the feedback

This sounds obvious and it’s the step I skipped for months. The tool says “done, here’s the revised version” and I read it fresh, as if it were a new draft — and I miss that it quietly ignored my third instruction, or “fixed” something by deleting the whole paragraph instead of rewriting it.

Now I check each revision against the feedback that produced it, point by point. Did the intro actually get shorter? Is the example still there? Did it keep the sentence I told it to keep? If something got dropped, I say so — “You kept the long intro. Cut the first two sentences.” — instead of letting it slide and fixing it myself later.

That loop is the whole game. Feedback that isn’t checked isn’t feedback; it’s a suggestion the tool is free to ignore. Checking closes the loop, and a closed loop is what turns three rounds of revision into a draft that’s actually better, not just different.

The honest truth

Most people blame the tool when the second draft disappoints them. But the tool is only ever as good as the feedback it gets, and “make it better” is not feedback. The writers who get the most out of AI aren’t the ones with the cleverest prompts — they’re the ones who learned to edit out loud: point at the exact sentence, show the fix, ask for one change at a time, protect what’s working, state the rule behind the correction, and check that it actually happened.

That takes longer than typing “make it better.” About five minutes longer, per draft. What you get back is drafts that improve on a curve instead of wandering sideways — and a tool that starts to feel less like a slot machine and more like a junior writer who’s learning how you like things done. That’s the version worth having.

Frequently Asked Questions

Why does the AI ignore some of my feedback?

Usually because the message contained several unrelated changes at once, and the weakest instructions got dropped. Send one change per message, in order of importance, and check each revision against the feedback before moving on.

How specific should my feedback be?

Specific enough to quote. Copy the exact sentence that’s wrong, name the problem with it, and show what you’d rather have — one replacement sentence in the style you want. An adjective alone (“punchier”, “clearer”) is not feedback the tool can act on.

Should I correct every mistake, or just the important ones?

Fix structure first, then tone, then details. Correcting every small thing in a section you might cut wastes a round. And when a problem repeats, stop correcting instances — state the rule once (“plain words, not Latinate ones”) so it stops recurring.

Does giving feedback actually make future drafts better?

Yes, within a conversation — and across them, if you keep feeding your edited versions back in. The tool learns your taste from what you keep and cut, not from what you say you like. Paste your final edit back with a note on what you changed and why, and the next first draft starts closer to your voice.

Try it on your draft. Repetition finder runs in your browser. Paste a revised draft in before you send feedback on it — it surfaces the words and phrases you lean on too often, so the next round of feedback can target the real repetition instead of the feeling of it.

Continue your workflow

Was this guide helpful?

Your selection stays on this page. To send feedback to the publication, use our contact form.

About this publication

From Text Craft Pro

Practical guidance on clear writing, thoughtful editing, and useful web content. Our guides use illustrative examples and original diagrams, with AI-assisted drafting and editing.

Read our editorial approach About Text Craft Pro

Published 11 November 2026 · Text Craft Pro · AI-assisted educational writing. Editorial policy · Suggest a correction

Scroll to Top