Giving feedback on a draft is a skill with a bad reputation, mostly because most people have only received bad versions of it. The bad versions share a pattern: a list of complaints with no priority, a rewrite of the writer's sentences in the reviewer's voice, and a general sense that the draft has failed without a clear account of what success would look like.
Good feedback is smaller, slower, and much more useful. It leaves the writer with a short list of decisions and the confidence to make them.
Read first, comment second
The single biggest error is commenting while reading. A comment made on paragraph two is often answered by paragraph six, and the writer then has to explain why the reviewer's objection is already handled. Both people lose time and the reviewer loses credibility.
Read the whole draft once without marking anything. Then wait a few minutes. Then write down the one or two things that stayed with you.
That first pass gives you something a line-by-line read cannot: the reader's experience. If you cannot remember the argument by the end, the draft has a structure problem, and no amount of comma-fixing will solve it.
Separate the levels of feedback
Drafts fail at four different levels, and feedback at one level should not be mixed with another.
The argument. Is the central claim clear, interesting, and supported? This is the highest-leverage level and the one most feedback skips.
The structure. Does each section do a job? Does the order make sense? Are there two introductions and no conclusion?
The paragraphs. Does each one have a point and a topic sentence? Does it stay on that point?
The sentences. Grammar, rhythm, word choice, punctuation.
Most reviewers jump straight to level four because it is the easiest to see. But a beautifully edited paragraph in an argument that does not work is a waste of everyone's time. Fix the highest broken level first, and say explicitly which level you are commenting on.
Observations, not orders
Compare these two comments.
You need to cut the second paragraph. It's repetitive and slows everything down.
I lost the thread in paragraph two — I wasn't sure whether it was still about the pricing decision or had moved to the rollout. Was that deliberate?
The first is an order. The writer can obey it or argue with it, and both options cost energy. The second is an observation about the reader's experience, and it gives the writer the information they need to make a decision.
The formula is: what I experienced, where I experienced it, and the question it raised. "I lost the thread here" is more useful than "this is confusing," because it describes an effect rather than delivering a verdict.
This does not mean feedback should be tentative. If a sentence is ungrammatical, say so plainly. Observations are for matters of judgment, where the writer's intent matters. Facts can be stated as facts.
Never rewrite their sentence for them
The temptation is strong. You see a flabby sentence, you know exactly how to fix it, and you type the better version. The writer now has a sentence in your voice, and their draft has become a patchwork of two styles.
Better: point at the problem and let them solve it.
Not: "The team was in agreement that the proposal should be rejected." → "The team rejected the proposal."
Instead: "This sentence is doing a lot of work to avoid saying 'rejected.' Is there a reason for the distance?"
The second version teaches the writer something they will use in the next draft. The first version teaches them that you are better at sentences, which is not the lesson you want to deliver.
There is one exception. If the writer has asked for a line edit, rewriting is the job. In that case, use track changes so the writer can see and accept each change, and keep the changes minimal. The track changes workflow covers how to do this without leaving a mess.
Say what works
Feedback that only lists problems distorts the draft. The writer cannot tell which parts to protect, so they may cut the best material while fixing the worst.
Name two or three things that work and why. "The opening example is doing exactly what it should: it makes the abstract claim concrete before you ask me to believe it." This is not praise for its own sake. It is information about what to keep.
The effect on the next draft is often larger than any correction, because writers under pressure tend to delete anything they are unsure about.
Match the feedback to the stage
A first draft needs structural feedback. A final draft needs line edits. Delivering the wrong kind is worse than delivering none.
If someone sends you a first draft and you return comma corrections, you have told them the argument is fine. If it is not, you have wasted the one round of attention they had. If someone sends you a final draft and you return "have you considered restructuring the whole thing," you have made the deadline impossible.
Ask what stage the draft is at before you read it. If the writer does not know, ask what they are worried about. The answer usually identifies the level.
Delivering the note
Keep the note short. Three to five points, ordered by importance, with the highest-level issue first. Attach line notes separately if you have them, and say clearly that they are secondary.
End with a next step. "I think the structure is the main thing; once that is settled the rest will be quick. Send me the next version and I will read it again." This tells the writer that the draft is fixable and that you are still in the conversation.
The measure of good feedback is not how much of it there was. It is whether the writer knows what to do next, and wants to do it.