Disclosure: This article may contain affiliate links. We earn a commission at no extra cost to you if you purchase through our links. We only recommend tools we genuinely believe in.
A client once sent me feedback on a landing page in four separate emails, a Slack thread, and a voice note recorded in a car. One comment said the headline was too long. Another, in a different thread, said the headline was the best part and should not change. A third asked for a completely different offer on the page. By the time I had collected it all, round two had quietly become round seven, the launch date had slipped two weeks, and nobody could tell me which version of the page was the current one.
The work was not the problem. The page was fine. The feedback process was the problem, and I had built it by accident: no limit on rounds, no single place to collect comments, no rule for what counts as a revision, and no moment when anyone says the work is done. Every client got a custom process, which meant every project got a new way to stall.
The fix was a system I now run the same way on every project: the rounds are agreed before the work starts, all feedback lands in one place by one deadline, AI sorts it into fix, clarify, or push back, and the round closes with a written sign-off. The revision loop is not a personality problem. It is a process problem, and a process problem can be fixed with a checklist.
This post builds on How I Write a Client Proposal With AI That Actually Closes, because the rounds are agreed in the proposal, not discovered mid-project, and it pairs with How I Handle Scope Creep Without Losing the Client, because most endless revision loops are scope creep wearing a feedback costume.
This is a playbook, not a results report. The round counts and timings below are a starting template to tune after your first few projects, not measured outcomes. Every client is different, so use your own numbers once you have them.
The Problem: Feedback Is Not the Enemy, Unstructured Feedback Is
Good feedback makes the work better. I have shipped better pages because a client caught a claim I could not prove, or a headline that only made sense to me. The problem is never that the client has opinions. The problem is that the opinions arrive with no structure, no deadline, and no end.
Five habits turn a healthy feedback round into an endless loop:
- Feedback arrives in six places. Email, Slack, a voice note, a comment on a screenshot, a text message, and a note from a call you were not on. Half the job becomes hunting down comments, and the other half becomes guessing which one is the latest.
- Every comment looks equal. "Fix the typo in the footer" sits next to "rethink the whole offer," and both get treated as one more small change. The typo takes two minutes. The rethink takes two days. Treating them the same is how a round eats a week.
- Nobody agreed what a round is. Is a round one batch of comments, or one comment at a time for a month? Without a definition, the client thinks they are still in round one, and you are on round five, and you are both right.
- Vague comments get literal fixes. "Make it pop" becomes a new color, a new font, and a new layout, and none of it is what the client meant, because the client meant "I am not sure this will convert." You cannot fix a feeling with a color change.
- There is no finish line. No one says the words "this is approved." The work just stops getting comments, and you ship it while wondering whether the client is happy or has simply stopped replying.
The fix is not to limit the client's voice. It is to give the voice a shape: a place, a deadline, a definition, and a sign-off. Structure does not silence feedback. It makes feedback usable.
The System in One Pass
The stack is deliberately boring, and it reuses the tools from the proposal and scope creep posts. No new platform is required to start, and no part of this system talks to the client on its own.
| Stage | Tool | What happens |
|---|---|---|
| Set the rounds | The proposal, written with AI from your scope notes | Two included rounds, defined in writing, priced beyond that |
| Collect | One shared doc or one form, one link, one deadline | Every comment lands in the same place, with a name and a date attached |
| Triage | Claude or ChatGPT, with the brief attached | Each comment sorted into fix, clarify, out of scope, or taste |
| Revise | Your normal workflow, in one sitting | All fixes from one round, done together, with a change log |
| Sign off | A short written approval, in the same thread | The client confirms the round, or names the next one, in writing |
Notice what AI does not do in this table: it does not decide whether a comment is right, it does not talk to the client, and it does not approve anything. It sorts the feedback against the brief you wrote. You make the judgment calls, and the client makes the final one.
Step 1: Put the Rounds in the Proposal, Before the Work Starts
The revision loop starts in the proposal, not in the feedback round. If the proposal says "unlimited revisions," you have sold an endless loop. If it says nothing about revisions, the client will assume the reasonable number is whatever they need, and you will assume it is two. Both assumptions feel fair until round four.
My proposal now names the number: two included feedback rounds on the first version of each deliverable. A round is one batch of comments, collected by one deadline, on one version of the work. Comments that arrive after the deadline wait for the next round. A new round after the two included ones is billed at the hourly rate named in the proposal, or folded into a change order, and the client knows that before the work starts.
Two practical details make this work in real life. First, the round count is per deliverable, not per project. A landing page and three ad variants are four deliverables, and each one gets its own two rounds, or the client will spend the rounds for the page on the ads and then ask for more on the page. Second, I say the number out loud on the kickoff call, in plain words, even though it is already in the proposal. "Two rounds on each deliverable, and I will tell you when we are on round two." Clients do not read proposals the way we do. Hearing it once, in a conversation, prevents the surprise later.
AI helps me draft this section of the proposal from my scope notes, but I read it before it goes out, because the round language is a promise. A vague promise here becomes a painful argument in week three.

Step 2: One Channel, One Deadline
When the first version is ready, I send one message with three things: a link to the work, a link to the feedback doc, and a deadline. The feedback doc is a simple shared document with one line per comment, and I ask for three things on every line: where on the page it applies, what should change, and why. The "why" is the part most people skip, and it is the part that saves the most time. "Change the headline" is a guess. "Change the headline because our buyers do not know what this product does yet" is a decision I can act on.
The deadline is a real date, usually three business days for a page, two for an ad set, and I name what happens if it passes: I move on to the next deliverable, and the comments join the next round. This is not a threat. It is how a calendar works, and clients respect it once they see it applied kindly and consistently. The first time I enforced a deadline, I expected a fight. What I got was a client who sent better feedback, faster, because the deadline made them read the page properly instead of skimming it between meetings.
One more rule: the person who gives feedback is the person who approved the brief, or one named person they delegate. Feedback by committee is how you get two comments that contradict each other on the same headline, and then a third comment asking why neither was applied. If the client wants three stakeholders to weigh in, they consolidate into one voice before the deadline. My job is to collect one round of feedback, not to referee a meeting I was not invited to.
Step 3: Triage Every Comment With AI, Then Decide Yourself
Once the deadline passes, I paste the feedback doc into Claude with the original brief attached and ask it to sort every comment into four buckets:
| Bucket | What it means | What I do |
|---|---|---|
| Fix | The comment is specific, matches the brief, and I can act on it | Do it in the revision pass |
| Clarify | The comment is vague, or I do not know what the client actually wants | Ask one question, before touching the work |
| Out of scope | The comment asks for something the proposal did not include | Quote it as a change order, or fold it into the next round with a price |
| Taste | The comment is a preference with no brief behind it | Apply it if it is cheap and harmless, or push back with a reason |
The triage takes about twenty minutes for a typical round, as a planning figure, and it changes the whole job. Before triage, a list of twenty comments feels like twenty tasks. After triage, it is usually eight fixes, four questions, three out-of-scope items, and five matters of taste. The work shrinks to its real size, and the hard part becomes visible: the four questions, not the twenty comments.
The clarify bucket is where most of the value lives. A vague comment like "make the offer clearer" could mean the price is hidden, the guarantee is missing, or the client simply does not like the wording. AI is good at flagging vagueness and suggesting the question to ask. It is bad at guessing which one the client meant. So I send the questions, in one message, before I change anything. One round of questions saves a round of revisions, and clients would rather answer a question than receive a fix they did not ask for.
The push back is the part people skip, and it is the part that earns trust. When a comment contradicts the brief, I say so, with the brief quoted, and I offer the alternative: "The brief says this page targets first-time buyers, and this comment asks for language only existing customers would understand. I can make the change, and it will cost us the new buyers, or we can keep the brief and test the other version as an ad. Which do you want?" Clients rarely argue with their own brief. They argue with a freelancer who silently does something they did not mean.
Step 4: Do the Revision Pass in One Sitting
With the triage done, the revision pass is straightforward: fix everything in the fix bucket, in one sitting, and keep a short change log as I go. One line per comment, what changed, and where. The change log is not bureaucracy. It is the proof, when the client reviews the new version, that every comment was seen, even the ones I pushed back on. A comment that disappears without a trace feels ignored. A comment with a line that says "pushed back, see the note" feels heard.
I do not revise comment by comment as they arrive. I wait for the deadline, triage the whole batch, and revise once. Revising live, as comments trickle in, is how a page ends up with three different headlines in one afternoon, because the third comment contradicted the first and I had already applied it. Batching is the whole trick. One version in, one version out, one change log.
If a fix is bigger than I expected, I say so before I do it. "This comment asks for a new section, which is about half a day of work. It is inside the round, so I will do it, and it means the page ships tomorrow instead of today." Surprises about timing are worse than surprises about scope, and a one-line warning prevents both.
Step 5: Close the Round With a Sign-Off, Not a Silence
When the revised version goes back, the message is short: what changed, what I pushed back on and why, what is still open, and one question: "Is this approved, or is there a round two?" The client replies in the same thread, in writing, and that reply is the record. Approved means approved. Round two means a new deadline, and the same process runs again.
This is the step that actually ends the loop. Without it, the work drifts: the client stops replying, I assume approval, the client assumes I am still working, and three weeks later someone asks why the page never launched. With a written sign-off, the round has an ending, and the ending is a fact both sides can point to.
If the client does not reply, I follow up once, after two business days, with the same question. If there is still no reply, I send a closing note: "I am treating this as approved and moving to launch. Tell me if anything is still open." Silence is not approval, but an unanswered closing note with a clear deadline is a decision, and I put it in writing so nobody has to guess.
What I Do When a Client Wants a Fourth Round
It happens, and it is not a failure. Some projects genuinely need a third or fourth round: a new stakeholder appears, the market shifts, the client learns something real from the first version. The system does not prevent extra rounds. It makes them visible and priced.
When round three arrives, I name it. "This is round three, and the proposal included two. I can do it at the hourly rate in the proposal, or we can fold it into a change order with the other items you mentioned." Most clients say yes without a blink, because the price was in the proposal they signed, and because I am not springing it on them. The ones who push back are usually the ones who never read the round language, and a calm reminder, quoting their own proposal, settles it.
The only version of this I refuse is the round that arrives as a trickle: one comment a day for three weeks, each one presented as "just a tiny tweak." A tiny tweak, twenty times, is a redesign, and it is also the most expensive way to revise anything, because every tweak reopens the file, the context, and the review. When the trickle starts, I consolidate: "Let me collect these into one round and we will do them together on Thursday." Clients are usually relieved. They were trickling because they felt guilty sending a long list, not because they wanted twenty rounds.
Lessons
- Agree the rounds before the work, not during it. The proposal is where the loop gets its limit, and the kickoff call is where the client hears it out loud.
- One channel and one deadline beat any tool. A shared doc with a date does more than a feedback platform nobody opens.
- Ask for the why, not just the what. "Change the headline because buyers do not know what this does yet" is a decision. "Change the headline" is a guess.
- Triage before you touch the work. Sorting twenty comments into fix, clarify, out of scope, and taste usually cuts the real work in half, and shows you the four questions that matter.
- Push back with the brief, not with your ego. Quoting the client's own brief is the fastest way to resolve a contradiction, and it costs nothing.
- Close every round in writing. An approved round is a fact both sides can point to. A silent round is a launch that never happens.
The endless revision loop was never about difficult clients. It was about a missing process: no limit, no channel, no deadline, no definition, and no ending. Add those five things, keep the client's voice at the center of it, and the feedback gets better at the same time the rounds get shorter. That is the whole system. Two rounds in the proposal, one doc, one deadline, a triage pass, a batch revision, and a written sign-off. It takes about twenty minutes of process per round, as a planning figure, and it has saved me more launch dates than any tool I own.
