Authors write responses to reviewers as though addressing the reviewers. Reviewers do read them, but the decision belongs to the editor, and the editor is reading your response to answer one question: has this been dealt with?
Writing for that reader changes almost everything about the document.
What the editor is doing with it
An editor with a revised manuscript and a response letter is looking for whether each substantive point has been addressed, whether what you say you changed is actually what changed, and whether the remaining disagreements are reasonable.
They are reading quickly, and they are cross-checking. If your response says a paragraph was revised, they will look at the paragraph. A response that claims more than the manuscript delivers is the single most damaging thing you can put in one of these documents, because it costs you the benefit of the doubt on every other point in the letter.
They are also judging tone, not out of delicacy but because tone is evidence. A letter that argues with everything suggests an author who has not engaged. A letter that concedes everything, including things that did not need conceding, suggests one who has not thought.
Structure
Use the mechanical structure. It is not elegant and it is exactly right.
Quote each comment in full, in the reviewer’s words, numbered. Respond underneath it. State what you changed and where, with a page or line reference, and quote the new text if it is short. Then move to the next one. Do not merge comments, do not reorder them, and do not respond to a theme in place of the points that make it up.
Two conventions are worth adopting. Make the reviewer’s text visually distinct from yours, by italics or indentation, so the document can be skimmed. And give every point a response even when the response is one line, because a comment with nothing under it reads as avoided whether or not it was.
Where two reviewers raise the same issue, answer it properly under the first and refer back under the second. Do not make the editor read the same paragraph twice, and do not leave the second one blank.
Answering well
Lead with what you did. The first sentence under a comment should say what changed, not why the comment was interesting. The editor is looking for the action.
Be specific about location. “We have clarified this” is not an answer. “We have revised the second paragraph of the Discussion (p. 14, lines 302 to 310) to state that the effect is correlative” is.
Do the experiment, or say clearly why not. When a reviewer asks for something you cannot do, say so plainly and give the real reason, then say what you have done instead. A reason an editor can evaluate is far better received than a sentence implying the request was unreasonable.
Weaken the claim when the data do not support it. This is frequently the fastest route through a review and the one authors most resist. If a reviewer says you have over-claimed, the cheapest fix is almost always to claim less. Papers are rarely rejected for being appropriately modest.
When a reviewer is wrong
Sometimes they are. They have misread a figure, missed a control that is in the supplement, or asked for something the paper explicitly addresses. This happens, and you are entitled to say so.
Do it without heat. Set out what the reviewer appears to have understood, state what the paper says, and point at where. Then, and this is the part that matters, ask yourself why they misread it. A reviewer who read the paper carefully and still got it wrong is evidence that the paper is unclear at that point, and the right response is usually to answer the point and also improve the passage. Editors notice that move and read it as a sign of good faith.
If you are declining to make a change, say so explicitly rather than obscuring it. A clear “we have not made this change, for the following reason” is respected. A response that appears to agree while changing nothing will be found, and will be read as evasion.
Tone, concretely
Write plainly. Thank the reviewers once, at the top, and then stop thanking them. A letter that opens every response with gratitude for an insightful comment reads as padding and makes the document longer to assess.
Never characterise a reviewer or their motives. Never speculate about who they are. If a review is genuinely unfair or appears to have a conflict, that is a matter to raise with the editor directly and separately, not in the response document.
Reviewers who disagree with each other
This is common and it is not your problem to resolve alone. When two reviewers want opposite things, say so explicitly, explain the choice you have made and why, and leave the editor to adjudicate. That is their job and they are expecting to do it. Quietly satisfying one and ignoring the other is worse, because it looks like you did not notice.
Before you send it
Read the response letter and the revised manuscript side by side and check every claim in the letter against the text. Every one. This is tedious and it catches the mistakes that matter most, because a response letter and a manuscript that disagree is the failure mode that costs you credibility rather than just a round of revision.
Then check you have answered every numbered point, including the small ones about references and figure labels. Unaddressed trivia is the most avoidable reason for an extra round, and an extra round is weeks.