Does a country label mean those ratings came from that country?

A country label does not make an app’s rating count local.

Google Play rating counts looked almost identical across countries. Apple’s usually looked very different. That changes what these numbers can tell us about local ASO performance.

Prepared 27 October 2026 · Observations: 19 September to 26 October 2026, UTC · 244,883 source records · 89,309 country comparisons · 1,561 store-specific app IDs

Google Play’s country-labelled rating counts behaved like a shared cross-country measure. Treating each country’s count as a separate local audience would be misleading. The data supports that warning strongly, although it cannot establish exactly how the store or collector produced the shared counts.

1. What we tested

Imagine an ASO report saying an app has 36 million ratings in the UK and another 36 million in the US. Are those two different groups of people? Or is the report showing almost the same broader count twice?

We compared the same app across two countries, in the same store, on the same date. The observations had to be no more than six hours apart, and both counts had to be at least 100. We called the counts a close match when they differed by no more than 1% of the larger count.

In the later period, the difference between stores was enormous:

The headline percentages give every app equal weight. First, we calculate each app’s share of matching country comparisons. Then we average those shares. An app appearing in many countries does not receive more weight just because it creates more rows. The interval measures variation when resampling the apps we observed, not uncertainty about the entire app market.

For a simple sense of scale, the average Google Play app matched across countries almost every time. The average Apple app almost never did. The later-period gap was 98.21 percentage points, with an app-resampling interval of 97.50 to 98.86 points.

2. WhatsApp: identical count, different star scores

On 19 October, WhatsApp’s Google Play records for Japan and South Korea were recorded just 7.512 seconds apart.

The count was exactly the same: 246,004,115. The star scores were different. That is a concrete example of why a country-specific star score does not make the accompanying count a count of local raters.

These two records do not identify the people behind those ratings. They do show why adding the two counts together would not be justified.

Spotify shows the contrast between stores

Spotify’s Google Play counts differed by only 1,002 out of roughly 36.4 million, even though the records were about five hours apart. Its Apple counts were roughly 7.0 million in the UK and 42.5 million in the US. These are illustrative examples selected after the analysis; the percentages above come from every eligible comparison.

3. The weekly increases also matched

Matching totals alone could be a weak clue. We therefore checked whether the same app’s count also changed by a similar amount in different countries over seven days.

Both countries needed observations at the start and end. We allowed at most six hours between country observations at each endpoint. For this comparison of increases, both counts had to rise by at least 10. We called the increases close when they differed by no more than 10% of the larger increase.

These percentages also give each app equal weight. The later Google Play interval was 96.73% to 98.58%; Apple’s was 4.47% to 8.30%. This is a secondary check, conditional on both counts increasing by at least 10.

Spotify’s weekly change, using the actual recorded counts

The Google Play increases were 36,346 and 36,871, a difference of about 1.4% of the larger increase. Apple’s were 17,063 and 112,268. The store examples use different available weeks, so they illustrate the pattern rather than serving as a precisely matched cross-store experiment.

We use “net increase” deliberately. A change in a published count is not a verified count of newly submitted ratings or new users. Counts can fall too: 176 of the 3,493 later Google Play weekly comparisons had a decrease in at least one country. Those windows remain in the evidence file but are excluded from the positive-increase comparison.

4. Why this matters for ASO reports

A country label can make a number look more local than it really is. An ASO analyst might see a rising rating count in a UK competitor report and conclude that the app is gaining momentum specifically in the UK. In this Google Play dataset, almost the same increase often appeared elsewhere.

For Apple, the pattern was very different: country counts rarely matched. That is consistent with local differences, but it does not turn rating growth into proof of local downloads, installs from search, or ASO success.

The useful finding: In this source, Google Play rating-count growth provides little evidence that growth belongs to one particular country. Country-specific ranking, a country-labelled metadata row and a country-specific rating count are three different things.

5. We checked whether the result depended on our choices

It appeared in both parts of the month

The later period repeats the earlier pattern. It is a chronological repeat within an existing dataset, not a new independent sample or an untouched holdout.

Tighter timing and smaller samples gave the same broad answer

When observations were within one hour, Google Play reached 99.88%. Restricting the comparison to the UK and US still gave 96.92%, while Apple was 0.00%. Taking just one comparison per app also preserved the difference.

“Close” does not always mean “identical”

Google Play counts were often slightly different. That can be consistent with different observation times, updates or caching, but this study does not establish the reason. Our main result concerns near equality, not a claim that every record is an exact duplicate.

There were genuine failures of the 1% rule. BlinkDrama’s UK count was 302,506 and its US count 305,587 on 11 October, about five hours apart. The difference was 1.008%, just outside the rule. In total, 197 of the 29,620 later Google Play comparisons failed it.

6. What is established, and what remains an explanation

Established in the records: Google Play rating counts were overwhelmingly similar across countries, and their weekly increases were usually similar too. Apple’s counts and increases usually differed much more. The result survived tighter timing, higher minimum counts, one comparison per app and a UK-US-only comparison.

The likely explanation: The Google Play count field behaves like a shared count spanning countries. This is an inference from the observed pattern. We cannot prove that every count is worldwide, identify which users are included, or rule out collector or cache behaviour as part of the explanation.

What the store documentation says: Google describes country- and device-specific displayed rating scores, and its developer Console offers country/region breakdowns. Apple describes territory-specific summary ratings. These documents give context; they do not independently certify the geographic scope of the specific count field analysed here.

This is a measurement audit with an ASO implication, not a claim to have discovered a new app-store algorithm. It does not test whether adding keywords, changing screenshots or improving a rating causes more downloads.

7. How the study was done

1. Use the preserved source. We analysed all 244,883 records in the seven-field AppOrbit metadata extract. It includes apps and games discovered through the source’s chart collection, including free and paid titles. It is not a random sample of all apps.
2. Select one recorded observation per day. For each store, app ID and country, take the latest actual metadata event on each UTC date. There were 233,282 daily keys. We excluded 2,578 with a missing or invalid latest count, leaving 230,704. We did not fill gaps with an older count. No conflicting latest-timestamp records were found.
3. Compare like with like within each store. Use the same app ID and date, different countries, at most six hours apart, and counts of at least 100. This produced 62,542 Google Play comparisons across 996 app IDs and 26,767 Apple comparisons across 565 app IDs. We did not assume IDs or app populations were interchangeable across stores.
4. Give each app equal weight. Average the match indicator within each app, then average across apps. Raw row-weighted results are also included. For the later period, those raw results were 99.33% for Google Play and 0.40% for Apple.
5. Repeat by date and test other rules. Split at 8 October, then check one-hour and twelve-hour timing, a minimum of 1,000 ratings, UK-US pairs, and one chronological comparison per app per period.
6. Test seven-day changes. Require matched country observations exactly seven UTC dates apart, with each country’s elapsed time between 156 and 180 hours. Within each period, retain the earliest available nonoverlapping windows per app and country pair. A later window must start after the previous one ends. There were 10,230 windows in total. The positive-increase check uses only windows where both counts rise by at least 10.

The limits that matter

• One collector and a short period. The source covers 38 UTC dates. It cannot establish how every ASO platform collects or labels ratings.
• Metadata events are selective. The source records changes, not a guaranteed daily census of every app. Missing events are not evidence of zero change. Our sample contains only actual eligible observations.
• Recording time is not necessarily fetch time. The format describes run-level collection, enrichment and request reuse. A small timestamp gap improves comparability but does not prove two independent live storefront responses.
• Store populations differ. Apple and Google samples have different app coverage. The size of the difference is descriptive, not a causal effect of changing stores.
• No private performance data. We have no developer Console exports, user locations, keyword impressions, conversion rates or acquisition attribution for these apps. We cannot confirm worldwide coverage or assign a count change to local ASO work.

8. Evidence and sources

The evidence download contains the source extract, analysis code, protocol, every comparison, weekly windows, selected examples, verification output and file checksums. It is embedded in this HTML file, so the report and evidence travel together.

Evidence ZIP: 21.7 MB. Requires Python, NumPy and pandas to rerun. No images or external scripts are needed to read this report.

1. AppOrbit source data, pinned commit
   Third-party observations. Preserved extract, format guide and license included in the evidence download. Current GitHub page retrieval was unavailable during this study; calculations used the previously preserved source.
2. Google: Making Ratings and Reviews better for users and developers
   Primary-source context: country and device-specific displayed rating scores. Does not establish the geographic scope of this collector's ratingsCount field.
3. Google: View and analyze your app's ratings and reviews
   Primary-source context: Console country/region breakdowns, counts and rating publication delays. We did not access developer Console data.
4. Apple: Ratings, reviews, and responses
   Primary-source context: territory-specific summary ratings. Does not independently validate every count in the extract.

Publication takeaway: An ASO report can be organised by country while a rating-count field still behaves like a shared cross-country total. In this dataset, that was overwhelmingly the case for Google Play. Using that count as proof of local app growth would overstate what the data tells us.

Experiment 7 · Prepared for Gabriel Machuret · Source attribution: AppOrbit. Database rights under ODbL 1.0; third-party store content retains its separate rights. Dates in the analysis are UTC.