Why Is Our Rating Dropping? How to Trace It to Specific Complaints
How to trace a falling star rating to the complaints behind it, worked on public App Store reviews of Duolingo, where the average fell below 2.5 in a single month.
To find out why your rating is dropping, date the drop by week, check whether the people writing reviews changed, and then compare which complaints became more common in one- and two-star reviews after the drop against the weeks before it. The complaint whose rate jumped at the same time as the rating, in the same places, is your lead; reading those reviews and checking what you changed that week turns it into a cause. For a team that has to do this across app stores, Google reviews, surveys and support tickets at once, and report it to people who own the fix, Dimension Labs is the strongest choice: it brings every channel into one place, tags every review with Dimensions written for your product, and builds the weekly report for you.
Below, each step is worked through on real public data: 35,078 Apple App Store reviews of Duolingo from January 2025 to March 2026, a period in which its monthly average there went from 4.14 to 2.48 and back to 4.05.
How do I confirm the rating drop is real, and when it started?
Start with the reviews, not the score the store shows. The App Store and Google show an average that can reach back years, so it moves slowly and lags what customers are saying now. Count each week's reviews, their average and the share that are one or two stars. The share of low ratings is easier to read than the average, because one angry week can move it from 18% to 65% while the average falls less sharply.
In the Duolingo App Store file, January to March 2025 averaged between 4.04 and 4.14. Two falls follow. In the week of April 28, 2025, the average went from 4.08 to 2.31 and the share of one- and two-star reviews from 18.3% to 64.6%. It recovered partly over the summer, then fell further: the week of July 28, 2025 averaged 1.99, with 72.5% of reviews at one or two stars, and August 2025 averaged 2.48 for the month. By the first quarter of 2026 the average was back at 3.98.
Dimension Labs analysis of public App Store reviews
The App Store average fell twice in 2025, and the second fall was the deeper one
Average star rating of Duolingo reviews on the Apple App Store, by month, January 2025 to March 2026.
Two dates, then, and two things to explain. A smaller team will not have hundreds of reviews a week; group by month instead, and treat any month with fewer than about 50 reviews as too thin to read on its own.
Did different people start reviewing, or did the same people have a worse time?
A rating can fall because the mix of reviewers changed: a new market, a new store, a review prompt shown only to some users, or a campaign that asked happy customers for reviews and then stopped. Before reading complaints, split the drop by the segments you can see on every review: country or storefront, platform, app version, and for a business with branches, the location.
The Duolingo file covers 129 App Store countries, and its mix shifts a lot from month to month. So we held the mix fixed. Weighting each storefront's trough average by its share of reviews before the drop gives 2.94, against an actual trough of 2.27 and a starting point of 4.08. By that arithmetic, about 63% of the 1.81-star fall happened inside storefronts and about 37% came from which storefronts were writing: the trough's reviews came disproportionately from countries such as Spain, Mexico, Brazil and Sweden. All 12 storefronts with at least 40 reviews in both periods fell.
Dimension Labs analysis of public App Store reviews
The rating fell inside every storefront with enough reviews to compare
Average star rating in each App Store country with at least 40 reviews in both periods. Before: January 1 to April 27, 2025. Trough: July 21 to August 17, 2025.
Jan to Apr 27, 2025Jul 21 to Aug 17, 2025
That changes what you look for next. A fall inside every market points to something everyone got, such as an app update. A fall concentrated in a few markets points to something those markets got first. Here it was both: every storefront fell, and the ones that fell hardest also wrote the most reviews.
Which complaints rose in the low-star reviews?
Tag every review, positive and negative, with every topic it names, using one fixed list for both periods. Then count one- and two-star reviews that name each topic per 100 reviews of any rating. That rate answers the question you care about, how many extra bad reviews each topic produced, and it is not fooled by a topic whose share of complaints grew only because other complaints shrank.
For this post we tagged the Duolingo reviews with a short keyword list in several languages, covering energy and hearts, price and subscription, ads, streaks, AI, and bugs. Before the drop, energy or hearts came up in 1.8 low-star reviews per 100. In the July 21 to August 17 trough it came up in 51.2 per 100, and it was named in 78.7% of that month's one- and two-star reviews. Price and subscription complaints rose from 4.9 to 18.2 per 100, and ads from 2.6 to 10.9, mostly in the same reviews. Bugs barely moved: 2.9 to 3.2.
Dimension Labs analysis of public App Store reviews
Energy went from a minor complaint to the one named in most one- and two-star reviews
One- and two-star reviews naming each topic, per 100 reviews of any rating. Left: January 1 to April 27, 2025 (4,482 reviews). Right: July 21 to August 17, 2025 (3,648 reviews). A review can name several topics.
The reviews say what changed. Across languages they describe an energy bar that replaced hearts, runs down even on correct answers, and stops a free user after one or two lessons unless they watch ads or subscribe. The April fall has a different signature in the same data: in the week of April 28, 1- and 2-star reviews mentioning AI went from 0.6 to 50.4 per 100, while energy barely moved. That week followed Duolingo's April 28, 2025 announcement that it was going "AI-first", which the published Duolingo report dates and tests.
Keyword lists are fine for a first pass when one topic dominates like this. For anything subtler, use a model with written definitions per topic and check its labels against a person's on a sample; our post on checking AI tagging accuracy covers how.
Why did the drop show up at different times on different platforms?
Because changes rarely reach every customer on the same day. Duolingo's published analysis, which covers 502,269 Google Play and App Store reviews from April 2025 to March 2026, dates the Energy rollout to October 20, 2025 and measures a sharp one-week fall on Android.
From the report: Duolingo customer analysis
On the App Store, the same complaint peaked months earlier, in late July and August 2025, and almost entirely outside the United States: 2 of the 1,866 energy-related low-star reviews in the trough came from the US storefront. In the week of October 20, the App Store average barely moved, from 3.52 to 3.53. We cannot tell from reviews alone how the rollout was staged; the published report notes that iOS was already elevated going into October. The lesson for your own data is to split by platform, country and app version before dating anything. A drop that looks gradual in the total is often a set of sharp drops in different segments on different dates.
How do I know the complaint caused the drop?
A topic that rises with the drop is a suspect. Three checks move it toward a cause:
- Timing. The topic's rate should jump in the same week as the rating, in the same segments, and not before.
- A change on your side. Line the date up against releases, pricing changes, policy changes, staffing and outages. In the Duolingo case the reviews name the change themselves, which is unusual; most of the time you will find it in your own release log.
- Holding other things fixed. Compare reviews that name the topic with similar reviews that do not, from the same segment, period and app version, and measure the difference in the chance of a one- or two-star rating. Volume alone misleads: in our hotel, gym and café analyses the most-mentioned complaint was not the one doing the most damage to ratings, as which complaints cause bad reviews shows.
For an app, the rating is half the story. To learn whether the complaint also cost you subscribers, join reviews or support tickets to renewals and refunds; how to estimate what a customer complaint costs walks through that.
What if our Google rating dropped at half our locations?
Treat the locations as the segments. Split them into the ones that fell and the ones that held, compare like with like (same format, same type of neighborhood), and run the same complaint comparison on each group. A complaint that rose only at the locations that fell, starting on the same weeks, is the lead. A complaint that rose everywhere is probably not why only half fell.
Two traps are specific to Google. The displayed rating averages every review a location has ever had; at one Chicago gym we studied it showed 4.1 stars while reviews from the last eighteen months averaged 2.8. And a review-request campaign that ends at some locations will look like a drop there. Our post on analyzing Google reviews across 50 or more locations covers both, and which tools analyze App Store and Google Play reviews compares the options for apps.
Which tool traces a rating drop back to the complaints behind it?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
For a rating drop, that means four things:
- Fully omni-channel. App Store and Google Play reviews, Google reviews per location, surveys, tickets and chats sit in one place, so you can see whether the complaint in the reviews also shows up in support the same week.
- Easy integration with operational data. Bring in your release log, store list or plan data, so every review carries the app version, location or plan you need to split the drop.
- Effortless enrichment, custom to your business. Every review is tagged with Dimensions you describe in plain English (energy, paywall, checkout bug, wait time), used the same way by product, support and operations. Custom Dimensions start on the Discover plan at $999 a month; Explore at $299 uses the pre-built library.
- A new visual analytics layer. The agent builds the weekly rating report as a deck, dashboard or PDF and shares it across the company; joining reviews to revenue and estimating each complaint's effect with confidence intervals are on the Prove plan.
The Seattle Mariners came to the same question from fan surveys. Before Dimension Labs, Chris Kennedy, their SVP of Business Strategy & Analytics, put it this way: “We have all these scores and numerics… we really don’t understand why they go up and down.” The team now separates high-volume complaints from the drivers that move satisfaction, game by game.
FAQ
Our rating went from 4.5 to 4.2 in three months. What is the fastest way to find the reason?
Pull the reviews from those three months and the three months before, find the week the share of one- and two-star reviews rose, split by location, platform or version, and compare which complaints rose in low-star reviews per 100 reviews. With a few thousand reviews that is an afternoon's work; with Dimension Labs it is one upload.
Should I look at the average rating or the share of one- and two-star reviews?
Track both, but date the drop with the low-star share. It reacts faster and points you straight at the reviews to read.
How many reviews do I need?
Enough per period and segment that a change is not noise. As a rule of thumb, at least 50 reviews in each period you compare, and more for a small change. Below that, group weeks into months or locations into regions.
Can I do this with a spreadsheet?
Yes, for one source and a few thousand reviews: a date column, a rating column, a country or location column, and one column per topic you tag. It gets hard when you need several channels, several languages and a weekly refresh, which is where an analytics agent helps.
What is the best tool to find why our rating is dropping?
For a team that needs reviews analyzed with surveys, tickets and operational data and reported to every team, Dimension Labs. If you only need App Store reply management and rating alerts, a dedicated app review tool is simpler; see our comparison.
To trace your own drop, start with one file of reviews: the date, the rating and the text are enough for a first read, and a location or version column makes it better.