Volume is the only reliable repair

A bad run is a denominator problem before it is anything else. Two poor reviews out of six deliveries is a visible pattern; two out of twenty is a rough fortnight, and nothing about the two reviews differs between those cases.

That is the whole shape of the recovery. You cannot remove the reviews, you cannot usefully argue with them, and the only lever you hold is how much other evidence sits around them.

Diagnose before you change anything

Do this first and do it coldly, because the instinct after bad feedback is to change five things at once and then learn nothing from the next fortnight.

Three causes, and they need completely different responses.

Variance. You did the same work you always do and drew two buyers who were never going to be satisfied. This happens, and it happens more at low volume, which is exactly why a bad run in a quiet month feels like a verdict. Response: none, beyond continuing.

A drift in the work. Something has slipped - turnaround, length, the amount of attention per job - usually because volume rose or a format got stale. This is the most common real cause and the easiest to miss, because it is gradual and your sense of your own standard adjusts with it. Consistency across jobs is the underlying subject, and a variable standard damages a reputation faster than a consistently modest one.

A structural mismatch. A tier priced or scoped in a way that attracts buyers who expect something else. The cheapest tier is the usual suspect, because it reliably attracts the most demanding buyers, and no amount of care fixes a listing that is promising the wrong thing.

The way to tell them apart is to reread the reviews for what they name. Variance produces complaints about the thing itself, which is taste. Drift produces complaints about turnaround, length or thoroughness, which are checkable. Mismatch produces complaints about scope, which is the buyer describing a job you never offered.

Fix the cause, then stop fixing

One change, then a fortnight of jobs, then look again.

The temptation is to overcorrect: longer deliverables, faster turnaround, a lower price, a rewritten profile, all at once. Overcorrection is expensive and it is also self-defeating, because a set of changes with a shared outcome tells you nothing about which one mattered, and you now have a more elaborate process to sustain than the one that was working before.

Do not drop your price. It is the most common response and it is nearly always wrong, since it worsens the buyer mix in the direction that caused the problem in the first place, and a price you cut after bad feedback is very hard to restore.

Then produce volume

Once the cause is addressed, the repair is arithmetic and there is no shortcut through it. Buying one is not a shortcut either: in the US, the FTC's final rule on fake reviews, announced in August 2024, prohibits incentives conditioned on reviews expressing a particular sentiment, and groundless threats or intimidation used to remove negative ones.

Fifteen ordinary deliveries do more for the listing than any single brilliant one, for two reasons that stack. The average moves, slowly. And the text moves, quickly, because the reviews a buyer reads first are the recent ones, and a specific bad review from two months ago sitting under eight recent specific good ones has stopped being the thing the page says.

That second effect is the one that matters, and it is why the counts and the recency dominate the average - the same argument as in what buyers actually infer from a score.

Take the jobs that are easiest to do well while you are doing this. Not cheap ones, familiar ones. The formats you are fastest at, the briefs that arrive clear. This is the one period where turning down an ambiguous job is straightforwardly correct even if you would normally take it.

Reply to at most one of them

A calm, short, factual reply under a bad review persuades every future reader. A defence does the opposite, and a defence is what most replies become by the third sentence.

Two lines, no dispute of the facts, one statement of what you would do differently. Then never mention it again, including in messages to other buyers. The longer treatment of the first bad review covers the wording, and the principle is that you are writing for the reader in six weeks rather than for the person who left it.

Do not reply to all of them. A run of replies reads as a person in an argument with their own listing.

What a bad run is not

It is not a signal to stop, and separating a bad fortnight from a real trend is a general skill in this trade rather than a review-specific one - a quiet month is data rather than a verdict, and the same discipline applies here.

It is also not evidence that the work was wrong. Buyers frequently rate against an expectation formed elsewhere, and one of the places it is formed is the free automated score they ran first; the honest account of what automated scoring establishes and where it is uncertain is a useful reference when someone marks you down for disagreeing with a number a model produced. The tools that generate those numbers are what most buyers have already used, so a mismatch between your judgement and theirs is a conversation to have in the deliverable rather than a mistake to accept. And a look at what buyers tend to write and why will usually place your two bad reviews in a distribution rather than in a spotlight.

If the run coincided with a platform change to how listings are ranked or displayed rather than anything you did, that is worth checking directly; Rate Cock's judge pages document how work is surfaced there, and a drop in bookings with no drop in quality is often the ranking rather than the record.

The last part of the recovery is not doing anything special once it is over. The run ends when the run ends, and the earners who take longest to recover are the ones still running the emergency process four months later.