Mirakle Research · August 6, 2026
The praise came back. The complaints mostly didn't.
For three days in July 2026 Shopify's App Store served fewer reviews than it held, and 143,831 of them left their listings. Most came back. This is about which ones did — and which apps, and which reviews, were left short.
Part two. The week our scraper thought Shopify deleted 148,172 reviews established what happened, and that Shopify called the reviews incorrectly unpublished.
The three days were even-handed on the way out. Every category of app lost between 10.3% and 17.1% of its shelf; every size of app lost about the same again. Nothing in our data suggests the outage picked anybody.
The recovery is a different picture. Of the reviews that vanished, a five-star had a 52.4% chance of coming back and a one-star had 30.9%.
Five-star reviews that vanished
52.4% back
Out of 128,307 removed.
One-star reviews that vanished
30.9% back
Out of 6,705 removed. 5,598 one- and two-star reviews are still missing.
A rate on its own means nothing
Half of everything that left is still missing, and by itself that number has no scale to be read against. Reviews leave Shopify listings all the time — merchants delete them, stores close, Shopify removes ones it decides are not genuine — and some share of those never come back either.
So the same measurement runs over two other populations from the same season. Ordinary attrition: every review we recorded leaving between 3 May and 6 July, 59.8% of which are still missing. And the review-integrity sweep Shopify announced on 6 July and ran through the following week, where the figure is 78.9%. Both of those are Shopify acting deliberately, and in neither does a one-star review fare differently from a five-star one.
All three lines dip in the middle: a three-star review is the one least likely to be permanently removed under any of them, which we do not have an explanation for. What differs is where each line ends relative to where it started. Two of them come back to about the same height. The third does not.
The sweep is the useful comparison, because it is the case where Shopify is known to be choosing. It removed 17,999 reviews and kept off 78.9% of them, and it did that to praise and criticism alike. Whatever put the reviews back after the accident did not.
Shops that close take their reviews with them
There is one explanation that would make all of this ordinary, and it is a good one. Shopify removes the reviews of shops that no longer exist. It does that automatically, all the time, without anybody deciding anything. If the merchants who write angry reviews are also the merchants who give up and close, their reviews would stay missing more often, and nobody would have treated a complaint differently at any point.
That explanation rests on two claims. Both can be checked.
-
When a shop closes, its reviews stay gone.
True, and dramatically so. Sorting accounts by evidence the three days never touched — a review of theirs somewhere else that the event left alone — the accounts that have gone dark lost 96.9% of what was taken from them and got almost none of it back.
Confirmed
-
But a closing shop does not read the stars on its way out.
For those same dark accounts, a one-star review and a five-star review fared within 0.7 points of each other. The mechanism takes the whole shelf. It has no way to prefer a complaint, and it doesn't.
So it cannot make the gap on its own
-
The merchants who complain are the ones who leave.
This is the load-bearing claim, and it is false. Take every merchant whose reviews the three days never touched, so that nothing here is contaminated by the thing being explained. Of those with a single review to their name who gave it one or two stars, 21.13% have since vanished from the store. Of those who gave four or five, 20.47%. Among merchants with several reviews the complainers are less likely to be gone: 2.99% against 6.3%.
Not true
So the closure story survives its first claim and fails its second. Shops closing is real, and it is most of what never came back. But the merchants it takes are not the ones who complained, so it cannot be why the complaints are what is missing.
The same merchants, three different weeks
There is a cleaner way to see it, which needs no reasoning at all.
Most merchants in our record have written exactly one review, ever. For them a closed shop and a kept-off review look identical from outside, which is what makes the closure story hard to argue with. So stop arguing with it, and watch that one group of merchants through all three ways a review gets taken down: an ordinary week, Shopify's own announced sweep, and the three days.
Same merchants, same measurement, same wait to see what came back. Twice, the rating on the review made almost no difference. Once, it made 70.8% against 41.5%.
It is not a cliff between angry and happy either — it runs down the whole scale. One star 71.7%, two 66.4%, three 55.8%, four 43.0%, five 41.5%. The margins of error on the two ends are nowhere near each other: the one-star figure sits between 69.8% and 73.5%, the five-star between 41.0% and 42.0%.
Everything else we could think of
Four more things could produce the same pattern without anyone treating a complaint differently.
-
It could be length.
Unhappy reviews are longer — the median kept-off review runs 181 characters against the store's 110 — and long reviews came back less at every rating. But the gap is inside every length band, including the 192 removed reviews carrying no text at all: 52.6% of the low-star ones are still missing against 40.9% of the five-star. Those reviews are a number and nothing else.
Doesn't explain it
-
It could be the app.
Perhaps apps that collect complaints were restored worse. Across the 1,846 apps that lost both a low-star and a five-star review, so each app is compared only with itself, it is 67.6% against 44.4%.
Doesn't explain it
-
It could be who wrote it.
How many apps an account has reviewed matters enormously — for accounts with a single review to their name 43.5% of their removed reviews are still missing, and for accounts that had reviewed 25+ apps it is 98.5%. It is the strongest single thing we can find. It is not this: inside every one of those bands, the low-star reviews came back less.
Real, and separate
-
It could be the review itself — its age, its language, its history.
An old review might be harder to restore than a new one; a review the developer had answered might be handled differently; one its author had edited might carry a more complicated record; one written in a non-Latin script might sit in a different pipeline. Nine slices in all. The gap runs from 15.6 to 23.0 points across them, and holding account status roughly halves it without ever closing it.
Doesn't explain it
Where the argument stops
One result cuts the other way, and it belongs here rather than in a footnote. 2,460 accounts lost both a low-star and a five-star review across the three days. If a complaint were being handled differently review by review, those accounts should often have got one back and not the other. Mostly they did not: 95.9% saw both reviews share a fate. Among the few that split it leans the expected way — 28 kept the complaint off and returned the praise, against 15 the other way about — but that is 43 accounts, and nothing should be built on it.
So whatever produced the gap works mostly at the level of a whole account rather than picking through one account's reviews one at a time. That is a real constraint on what could have caused it, and it is why the effect shows most sharply among merchants who only ever wrote one review — the case where an account and a review are the same thing.
And the standing limit: we can see that a review is missing, not why. Shopify has published nothing per review. All of the above is about which reviews came back and none of it is about anybody's intent — an automated queue that gave up sooner on some records than others would produce these numbers just as well as a decision would.
Everyone lost the same. Not everyone got it back
Split the three days by app size and the two halves of the story sit on one chart. What left is a flat line: an app with under ten reviews lost 14.7% of its shelf, an app with over five thousand lost 12.7%. What came back is not flat at all.
The smallest apps got back roughly a quarter of what they lost. For an app with eleven reviews that is not a rounding difference; it is most of the evidence a merchant has to go on.
The same split by category says the same thing twice. Across the 63 categories large enough to report, the share of the shelf that left runs from 10.3% to 17.1% — a tight band, and the reason we do not think the outage picked anybody. The share that never came back runs from 28.4% to 74.8%.
At the bad end of that second row: Product feeds at 74.8%, then print on demand, dropshipping and store data importers. At the good end, Event booking at 28.4%, with pre-orders, wholesale, subscriptions and accounting alongside it. The categories that got their reviews back are the ones that sell tools to businesses that already exist.
What the missing reviews are
All of that is shape. The reviews themselves are readable, and they do not read like a random half.
The pile that never came back is longer than the store — 181 characters at the median against 110 — and it talks about different things. 2.47% of them mention dropshipping, suppliers or sourcing, against 0.42% of the store. 0.65% contain a link, against 0.13%.
Two registers sit in that list at once. One is a trade: dropshipping, suppliers, fulfilment, the vocabulary of people selling other people's products. The other is a tone — game changer, stands apart, genuinely, shoutout, dedication — the sound of a review written to be quoted. We can say those words are over-represented. We cannot say the reviews are not genuine, and the pile also contains this:
Those are the reviews a merchant goes looking for. Not the star average — the one that says what broke, on which plan, and how support answered. The first of them is a warning that an app's five-star reviews look fake, and it is one of the 5,598 low-star reviews that left during the three days and have not come back.
What came back
The returned pile is the easiest of the three to describe, because there is nothing to describe. On every marker we measured — length, rating mix, share with no text, links, vocabulary — it is the store. Median length 123 characters against the store's 110. It is what a Shopify review is.
And 42,099 that arrived instead
While the reviews were leaving, others appeared that we had never recorded on any listing, on apps we had been scraping for months. The median one had been written 804 days before we first saw it. The oldest are from 2014.
There is no single month behind them — it is a slope, thickest around reviews written two years before and thinning in both directions, which is what an accumulation looks like rather than one earlier event being undone.
They are also, by a distance, the plainest writing in this report. Shortest at the median (99 characters), 16.2% with no text at all, and when there is text it is usually a first name and a thank you.
What we are left with
An accident that treated everybody the same, and a recovery that did not. Small apps got back half as much of their shelf as large ones. Accounts with a long reviewing history got back almost nothing. Most of what is still missing is accounts that left Shopify altogether, which is ordinary and reads nobody's stars. What is left after that is a one-star review being 1.7 times less likely to return than a five-star one — a gap that survives length, the app, the reviewer's history, the review's age, whether the developer replied, whether it was edited, what script it is written in, whether the account still exists, and reviews consisting of nothing but a star.
Neither of the processes Shopify runs on purpose looks like this. Its announced integrity sweep took 78.9% of what it touched and took it evenly; its automatic clearing of closed shops takes 96.9% and takes it evenly. Whatever handed the reviews back in July was not even, and the 5,598 complaints still missing are the part of a listing that is hardest to replace: nobody writes them twice.
Every figure here comes from the committed snapshot behind this page, which you can read at /data. The quoted reviews are copied by hand and stripped of app names, developer names, support agents and store URLs; the extractor pulls no review text out of the database.