Most teams read one-star reviews the same way: scroll, wince, remember the angriest three, and file a ticket. That gives you a list of anecdotes. It doesn't tell you which problem costs you the most stars.
This is the method we used to sort 29,785 one-star reviews from 274 big-brand apps into themes. It needs a spreadsheet or a short script. No AI model is required, and you can check every step.
Step 1: pick a fixed window and write it down
The App Store and Google Play show the newest reviews first. "The latest 300 reviews" sounds like a fixed sample, but it isn't. For a busy app it covers a few weeks. For a quiet one it can cover years: in our dating app reports, Tinder's latest 300 App Store reviews started in August 2026, while Match's went back to December 2020.
So choose a date range first, then take every review inside it. In our study that was 25 Oct 2025 to 24 Oct 2026. If you compare yourself with a competitor, use the same range for both.
Step 2: count shares, not reviews
A theme with 200 reviews in a big app and a theme with 20 in a small one can be the same problem. Divide by the number of one-star reviews in the window and compare percentages.
Set a floor, too. We only ranked an app if it had at least 30 one-star reviews in the year, and only reported a sector or country with at least 100. Below that, one bad week can swing the share by ten points.
Step 3: define each theme as a pattern you can read
We used one regular expression per theme, matched against the review title and text. Here are three of the ten, exactly as we ran them:
-
Login & account access:
\blog ?in\b|\blogin|\bsign(?:ed|ing)? ?(?:in|on)\b|\bpassword|verification code|\botp\b|\b2fa\b|two[- ]factor|face ?id|touch ?id|fingerprint|locked out(shortened here; the full pattern is in the study) -
Broke after an update:
(?:since|after) (?:the |this |last |latest |recent |new |an? )?(?:update|upgrade)|(?:new|latest|recent|last) update -
Ads:
\bads\b|\badvert|\bcommercials?\b
A pattern is crude, and that is the point. Anyone on the team can read it, argue about it and change it. If someone says "we have a crash problem", you can count it in a minute instead of debating it.
A review can match several themes. A review that says "since the update I can't log in" counts for both, which is what you want.
Step 4: check a sample of matches by hand
Patterns catch things they shouldn't. Before trusting any number, pull a few random matches per theme and read them.
We checked 6 random matches for each of our 10 themes. 55 of the 60 were on topic. Two of the misses show the kind of error to expect:
- "crashed" in a review about a taxi driver, not the app
- "checkout" in a complaint about an upsell, not a failed payment
If more than one in five matches for a theme is wrong, tighten the pattern before you report it.
Step 5: look hard at what matched nothing
In our study, 59.4% of one-star reviews (17,699 of 29,785) matched none of the ten themes.
That is normal, and it is worth reading. Much of it is short ("useless", "worst app") and carries no signal. But some of it is a theme you haven't named yet. If you see the same complaint five times in the unmatched pile, add a pattern for it.
Also expect lower match rates outside English. Our patterns are English, so we only used English-language storefronts. Even then, Indian reviews often mix English and Hindi, which lowers the match rate there.
Step 6: split by store and by country
Averages hide the useful part. Three examples from the same data:
- By country. Customer service was 21.3% of one-star reviews in India, against 8.0% in the US and 2.5% in Ireland. Login was 2.8% in India and 14.5% in Canada.
- By store. 44.0% of App Store reviews in the sample were one-star, against 29.4% on Google Play. The top theme was login on both.
- By sector. "Broke after an update" was 3.3% overall but 10.3% in smart home and security apps. Ads were 3.6% overall and 24.7% in news and weather.
Be careful with country comparisons. They mix in which apps people use. India's sample leans towards shopping apps, and shopping apps attract service complaints.
Step 7: separate app problems from business problems
The three biggest themes in our study were login (9.5%), customer service (9.1%) and charges, fees and refunds (8.4%). Crashes came fifth at 4.6%.
Only one of the top three is clearly an engineering problem. Service and billing complaints are about refunds, hold times and surprise charges. The app is just where people go to say it. Route them to the team that can fix them, or they will stay in your reviews no matter how stable the build is.
Step 8: rank by share, then read the top reviews in full
Now sort themes by share and read the 10 to 20 most recent reviews for the top two. This is where the anecdotes become useful, because you know they represent a real share of the pain. Write down the specific failure each review describes ("code never arrives", "Face ID stopped after update") and count those too.
Mistakes this method catches
- "Crashes are our biggest problem." Often false. In our sample, login and service each beat crashes by about two to one.
- "Reviews got worse this month." Check the window first. A quiet app's "latest 300" may span years.
- "Our competitor gets the same complaints." Compare shares in the same date range, or the comparison means nothing.
- "The AI summary says users want X." Maybe. A pattern count lets you check how many reviews actually say it.
What you need
- An export of your reviews with date, rating, country and text. Both stores give developers their reviews through the console and an API. For competitors' iOS apps, Apple's public review feed works.
- A list of 8 to 12 themes as patterns. You can start from the ten we published.
- An hour for the first pass, then minutes each month after.
The full pattern list, the hand check and the per-app tables are in the one-star review study. Copy the patterns freely; the data is CC BY 4.0.
Short on time? Revunda's free check reads your app's latest reviews and sorts the 1 to 3 star ones into themes for you. No signup.
Top comments (0)