ReviewFold
Measurement · 24 apps, one afternoon
Two hundred reviews can mean eight hours or two years
“Sample the latest 200 reviews” sounds like a quantity. It is also a time window, and the window is whatever 200 reviews happen to cover. We measured it for 24 top US App Store apps on 7 November 2026. The range was 0.33 days to 668 days — a 2,057× spread.
01 · The measurement
We asked one question of every app.
For each app: take the newest 200 public reviews in the US storefront, in date order. How far back does review number 200 reach?
- Same count, every app
200 reviews each, taken two pages at a time from the same public endpoint. No app got a bigger sample than another.
- Same sort order
Newest first. This is the sort a person sees when they open “most recent” in the store, and the one most export tools default to.
- One afternoon
All 24 collections ran within about two minutes of each other, so the fast-moving apps reflect the same moment.
The count stayed fixed while the window moved. That is the whole finding: a fixed number of reviews is not a fixed amount of time.
02 · The spread
Eight hours at the top, twenty-two months at the bottom.
Sorted by how much time the same 200 reviews cover. Read the middle column first; the last one explains the rest.
| App | Window for 200 reviews | Reviews per day | 1–2 star share |
|---|---|---|---|
| Spotify | 0.33 days | 616 | 17.0% |
| YouTube | 0.36 days | 563 | 56.5% |
| Duolingo | 0.70 days | 286 | 16.0% |
| TikTok | 1.13 days | 177 | 41.0% |
| Uber | 2.09 days | 96 | 28.5% |
| 2.15 days | 93 | 59.0% | |
| Amazon Shopping | 3.67 days | 55 | 70.5% |
| Netflix | 4.20 days | 48 | 27.0% |
| 5.09 days | 39 | 30.5% | |
| Google Maps | 5.53 days | 36 | 56.0% |
| 8.06 days | 25 | 79.0% | |
| Discord | 8.15 days | 25 | 48.5% |
| Gmail | 13.6 days | 15 | 81.5% |
| MyFitnessPal | 18.0 days | 11 | 47.0% |
| Airbnb | 20.7 days | 9.7 | 96.5% |
| Calm | 23.6 days | 8.5 | 22.5% |
| Slack | 27.0 days | 7.4 | 34.5% |
| Strava | 51.6 days | 3.9 | 38.0% |
| Forest | 57.5 days | 3.5 | 67.0% |
| Notion | 141.5 days | 1.4 | 46.5% |
| Headspace | 176.6 days | 1.1 | 71.5% |
| Todoist | 265.2 days | 0.75 | 24.0% |
| Trello | 421.2 days | 0.47 | 51.5% |
| Things 3 | 668.5 days | 0.30 | 16.5% |
Median window across the 24 apps: 10.9 days. Ten apps fit 200 reviews into under a week; two needed more than a year. The review volume, not the count, sets the window.
03 · Consequences
The same ranking means different things at the two ends.
A theme table usually lists groups by how many reviews mention them. That number is a count inside a window — so the window decides what the ranking is about.
- At eight hours, you are reading a mood
Spotify’s 200 reviews are one day of App Store traffic. Whatever tops that list can vanish by the weekend, because it is news, a bad release, or a single outage — not a settled product problem.
- At two years, you are reading an archive
Things 3’s 200 reviews reach back to January 2025. A bug reported then and fixed in March still sits in the same list as a complaint from last week, with nothing in the count to separate them.
- In between, the count is silently a rate
For a mid-volume app at 25 reviews a day, 200 reviews is eight days. Whether that is “recent” depends entirely on how fast the product changes — which the count cannot tell you.
This is the same arithmetic problem as ranking praise and complaints together, one layer up. Complaints scatter across many phrasings and praise pools into a few, so a count-based ranking flatters a high-rated app. A count-based ranking over a sample with no stated window compares eight hours to two years and calls it a leaderboard.
04 · What we do about it
Three rules, and they are cheap.
None of them requires a bigger sample. They require writing down something the sample already knows.
- State the window, not just the count
“200 reviews, 30 October – 7 November 2026” is checkable. “200 recent reviews” is not, because two people will read it as two different periods.
- Treat anything under a week as sentiment, not backlog
A short window is excellent for “what broke this week” and poor for “what should we build next”. Label it for the first job and do not promote it to the second.
- For slow apps, split by version before counting
When the window runs past a release, group reviews by app version first. Otherwise you are summing problems from two different products.
A fourth rule, if you have the data: report both a 30-day and a 90-day window and show where the ranking changes. The change is usually more informative than either list.
Boundaries
What this data cannot tell you.
We would rather state these than have them found later.
- One storefrontUS App Store only. Other regions can differ in both volume and tone.
- One sort orderNewest first, not Apple’s default ranking and not “most helpful”.
- One snapshotCollected 7 November 2026. For the fastest apps, the window is no longer reproducible.
- No review text in the datasetWe stored IDs, dates, ratings and versions. Theme-level claims cannot be checked against it.
- Not a random sampleThe 24 apps were chosen for recognisability. This is not an estimate of the App Store as a whole.
05 · Reproduce it
Method, and the table as a file.
Everything below came from public App Store responses fetched with the script linked at the end of this section.
- Resolve the app
The public iTunes Search API returns the numeric app ID for each app name. No scraping of the web page.
- Fetch the reviews
The storefront’s own review endpoint, 100 rows per request, two requests per app, newest first, US storefront. Requests were spaced by 0.6–0.9 seconds.
- Compute the window
Newest date minus oldest date across the 200 rows, plus reviews per day and the 1–2 star share over the same rows.
The endpoint is the one the App Store front end uses. It is undocumented and has no availability guarantee, so this is a bounded public sample, not a full export — the same framing ReviewFold uses when it describes what it collected.
Dataset: app-review-sampling-2026-11.csv — 24 rows, one per app, with app ID, window in days, reviews per day, 1–2 star share, date range and total review count.
Script: app-review-sampling-2026-11.mjs — Node 18 or newer, no dependencies, writes raw.json. It reads only public Apple endpoints, so you can rerun it and compare windows.
Both files are free to reuse with attribution to ReviewFold.
Current support, stated plainly
- Apple App Store: US and Canada. Google Play: US.
- Up to 200 recent public reviews per report.
- One app and one region per report; no cross-app comparison.
- A free preview shows themes and counts; the paid report adds review excerpts, PDF and CSV.
- No claim that a sample window is representative of an app’s whole history.
If you would rather not compute the window yourself
What the report states today.
Every ReviewFold report names the storefront, the date the sample was collected, and says plainly that period coverage is not guaranteed. This page exists because that is the minimum, not the finish line: printing the window itself is the next thing worth doing.