Can growing rating counts predict which apps will climb?

ASO RESEARCH · EXPERIMENT 3 · 26 OCTOBER 2026

Can faster rating growth tell us which apps will climb the charts?

We tested 11,655 later app-chart observations. Adding recent rating growth barely changed our ability to predict a climb in either app store.

The ASO theory, in plain English

An app’s star rating tells you how people have rated it. Growth in its rating count might tell you that more people are using it now. Our theory was that this recent activity would help identify apps about to climb the charts.

Illustration: imagine two apps sitting around #50. One has 4.8 stars and its rating count barely changed. The other has 4.3 stars, but its count grew from 10,000 to 11,000 in roughly a week. Would the second app be more likely to climb over the following week?

That example explains the question. It is not a measured result.

Why this matters for ASO

ASO research involves deciding which competing apps deserve attention. If rating growth predicted a coming climb, an ASO team could spot emerging competitors before their chart position made them obvious.

This study tests that forecasting idea. It does not measure whether improving your own app’s ratings increases downloads. It also uses top-free charts, which are different from rankings for searches such as “budget planner” or “fitness tracker”.

What we tested

1. We looked at apps between #6 and #100 in complete free-app charts.
2. We measured how their rating counts had changed over approximately the previous week.
3. We used only information available before making each prediction.
4. We checked whether each app climbed at least five places seven calendar days later.

For example, moving from #50 to #43 counts as a climb. Moving from #50 to #48 does not. An app that disappears from a complete top-100 chart does not count as a climb.

We built the prediction models using September 26 to October 3 observations. We tested them on October 11 to October 19 observations, with later outcomes available through October 26. Dates refer to each market’s local calendar. Every training outcome was observed before the first test prediction.

An observation is one app in one country on one date. The same app can appear many times. We gave each app ID equal total weight within its store. The 643 Apple IDs and 573 Google Play IDs are not necessarily 1,216 different app brands.

The results

All four models below already knew the app’s chart position, previous week’s chart movement, total rating count and country. We then added stars, recent rating growth, or both.

What these numbers mean: they measure how well the model separates apps that later climbed from apps that did not. A score of 50 means chance-level ordering; 100 means perfect ordering. A score near 64 means it puts the future climber ahead in roughly 64 out of 100 weighted climber-versus-non-climber comparisons. It does not mean 64% of all predictions were correct. The technical name is ROC AUC, multiplied by 100 here.

Apple: adding rating growth produced 63.57, compared with 63.49 when adding stars. That is a difference of just 0.09 points out of 100.

Google Play: adding rating growth produced 58.92, compared with 58.87 when adding stars. That is a difference of just 0.05 points.

Before examining the results, we required rating growth to improve this score by at least two points over both alternatives, in both stores, with further checks. Neither store came close to that bar.

Real apps show why rating growth is not a simple signal

These are the three fastest-growing rating counts among eligible US apps in each store on October 11. The last column shows their chart position on October 18.

In the same slice, some apps climbed, others fell and one left the top 100. These examples illustrate the result; the full test includes all eligible observations, not just these apps.

The count endpoints come from the latest available records before the October 4 and October 11 chart captures. Records could be up to 72 hours old, and the measured count interval had to be five to nine days. Names are the names recorded in the source data.

Did the answer change when we checked it differently?

We repeated the comparison using just the first later prediction for each app, so frequently observed apps could not dominate. We also rebuilt the models using rating records no more than 24 hours old.

The numbers above are rating growth’s advantage over stars, in score points. Positive means growth performed better; negative means stars performed better. All differences stayed small. Comparing absolute daily count gains instead of proportional growth did not improve the main results either.

What this means for an ASO team

A fast-growing rating count was not a useful upgrade to this short-term chart forecasting method. It might still be worth observing as part of competitor research, but these results do not justify treating it as evidence that an app is about to climb.

We also cannot turn this result into “ratings do not matter for ASO”. Apple states that ratings and reviews can influence App Store search ranking, and that ratings appear in search results and on product pages. That is a different question from predicting next week’s top-free chart movement.

One possible explanation is that extra ratings follow activity that has already happened, rather than predict what happens next. Advertising, seasonal demand and changes to an app’s rating prompt could also affect the pattern. We did not measure those factors, so these remain explanations to investigate, not findings.

What this study can and cannot establish

We tested a specific prediction method on a short, selected sample of already-charting apps. We found very little extra predictive value from either stars or recent rating growth. We did not run a randomized test, observe private download counts, or measure keyword rankings.

The later dates were withheld from model fitting, but many apps appeared in both periods. This tests later performance for a mixture of familiar and newly observed apps, not forecasting only brand-new apps. The period covers nine test dates, and observations across dates and countries are related.

Apple’s broad apps chart includes games; Google Play’s apps chart does not. We therefore kept the stores separate. We also do not assume that Google’s returned rating count measures activity only within the chart’s country.

Verdict: the proposed theory was not supported. The result is useful as a narrow test of a plausible ASO shortcut, but it should not be published as proof that rating growth predicts app success.

Sources and reproducibility

Data: AppOrbit, pinned source commit 9621042dde5ad42193bc84c8b54e5bf9b2d4e9eb. Archived collection covers September 19 to October 26, 2026. The evidence bundle includes all chart rows, the snapshot index and all metadata rows with the seven fields needed for this experiment, plus analysis code, protocol, predictions and verification.

Platform context: Apple: App Store search; Apple: ratings and reviews; Google: country and device differences in ratings. Platform documentation provides context; it is not evidence that this prediction theory works.

AppOrbit-owned database rights are licensed under ODbL 1.0. Its license does not grant rights to third-party app-store content. The original license is included in the evidence bundle. This is study 3 in an exploratory series; study 1 did not support reactive listing edits, while study 2 measured differences between countries’ app charts. This result is retained regardless of its direction.

The evidence download is built into this HTML file. Open the saved report in a web browser to use the download button. No external images, scripts or fonts are required.