TL;DR: We ran 1,163,767 recent App Store reviews from 5,486 apps through Jev, put an identical list of questions to each one, and paid $65.60 for the lot. Pricing and paywalls come up in 195,188 reviews, more than every bug report combined. About a quarter of four-star reviews are praise with no complaint attached. One review in 22 describes a charge the user didn't expect or a subscription they couldn't cancel, and that share swings tenfold between Games and Food & Drink. When an update goes wrong, the review feed shows the exact version that did it.
The reviews come from the same data our App Store API and MCP server serve. Analysis ran September 19–21, 2026.
Key numbers
- 1,163,767 reviews, 5,486 apps, 23 categories, $65.60.
- 26% of four-star reviews are plain praise. 52% are mixed and 17% are negative.
- 49% of three-star reviews are outright negative, and another 44% are mixed.
- About 5% of five-star reviews have negative text.
- Pricing or paywalls appear in 18% of all reviews and 35% of negative ones.
- 20.5% of game reviews complain about ads.
- 4.6% of reviews (49,711) describe an unexpected charge or a blocked cancellation: 0.8% in Games, 8.3% in Food & Drink.
- In 677 apps (12%), at least 8% of recent reviews describe one.
- Reviews about customer support mention leaving 13.6% of the time. Reviews about crashes: 5.6%.
- 3.5% of reviews blame a specific update. In 392 apps it's more than one in ten.
- News is the angriest category (52.6% negative). Books is the calmest (28.3%).
- 27% of reviews contain no concrete information at all.
Star ratings are honest at the ends and vague in the middle
We compared the stars on each review with what the text says. On a 10% sample of apps (107,501 reviews):
| Stars | Positive text | Negative text | Mixed |
|---|---|---|---|
| 1★ | 0.5% | 94.5% | 3.8% |
| 2★ | 1.2% | 77.8% | 19.2% |
| 3★ | 4.1% | 48.5% | 43.7% |
| 4★ | 26.3% | 16.9% | 51.9% |
| 5★ | 81.5% | 5.4% | 8.4% |
One star means what you think it means. Four stars doesn't. Seven in ten four-star reviews contain a complaint, and the typical one reads like this Baby Tracker review from January, titled "Would give 5 stars but HATE NEW UPDATE." The author still gave four.
If you read your average rating as a satisfaction score, the four-star pile is where it misleads you. Those users like the app and are telling you what's about to cost you the fifth star.
The five-star complaints are stranger. We read 45 of them. About one in ten was the model misreading short praise, which puts the real share near 5%. The rest are ordinary angry reviews with the wrong star count: titles like "Scam DO NOT DOWNLOAD" and, on a five-star review, "1 star." Nothing in the text suggests people do it on purpose, and nothing explains the five stars either.
The top complaint is the price
Every review was sorted into one kind:
| Kind | Share | Avg severity (0–3) |
|---|---|---|
| Praise | 41.3% | 0.11 |
| Design complaint | 24.1% | 1.62 |
| Bug report | 17.4% | 2.22 |
| Billing issue | 5.4% | 2.59 |
| Feature request | 5.2% | 1.05 |
| Other | 6.6% | 0.83 |
"Design complaint" means the app works as built and the user dislikes how it was built: the paywall, the ads, the limits, the redesign. There are more of those than bug reports.
Then we tagged topics. A review can have several, or none.
By number of reviews: pricing / paywall 195,188, ads 98,672, content quality 94,745, usability / design 62,921, crashes / stability 43,273, customer support 29,644, usage limits 26,700, device compatibility 24,711, sync / data loss 24,314, login / account 21,187, notifications 17,404, privacy 11,442.
Pricing is in 195,188 reviews, including 35% of all negative ones. All bug reports together are 186,464. Crashes, the thing engineering teams fear most in a review feed, are less than a quarter of the pricing pile.
Billing issues are the smallest kind and the most severe: 2.59 out of 3, and the most negative language of any group: average polarity −0.83, against −0.57 for bug reports.
Users forgive crashes more than silence
Every review was also checked for whether the author says they are leaving, cancelling, or switching to something else. 3.9% do. By topic:
Customer support 13.6%, pricing / paywall 10.7%, privacy 10.3%, notifications 9.5%, ads 9.0%, sync / data loss 8.6%, login / account 8.0%, usage limits 6.8%, usability / design 6.5%, crashes / stability 5.6%, device compatibility 5.2%, content quality 4.0%.
Login failures, crashes, and lost data score highest on severity (2.5–2.6 out of 3), and none of them comes close to support on leaving. Crashes are near the bottom. Support complaints are rated less severe than any of the three (2.3), and their authors are the most likely to leave.
Every category has its own complaint
Shares of each category's reviews:
| Category | Negative | Surprise charges | Ads | Blames an update |
|---|---|---|---|---|
| Games | 38.4% | 0.8% | 20.5% | 2.8% |
| Health & Fitness | 35.5% | 7.2% | 2.4% | 3.8% |
| Entertainment | 40.7% | 3.8% | 12.8% | 2.2% |
| Photo & Video | 41.3% | 6.5% | 3.6% | 3.9% |
| Productivity | 37.7% | 3.2% | 3.4% | 4.6% |
| Utilities | 44.7% | 5.5% | 5.7% | 2.7% |
| Education | 33.6% | 7.2% | 3.0% | 1.5% |
| Lifestyle | 44.4% | 5.9% | 3.3% | 3.0% |
| Social Networking | 51.0% | 2.5% | 4.3% | 3.9% |
| Music | 39.9% | 4.8% | 8.4% | 2.7% |
| Graphics & Design | 43.7% | 5.5% | 9.5% | 2.5% |
| Reference | 30.1% | 5.2% | 4.9% | 1.9% |
| Business | 39.2% | 5.8% | 1.1% | 3.5% |
| Sports | 48.6% | 3.2% | 6.4% | 5.5% |
| Finance | 41.3% | 5.5% | 3.4% | 3.4% |
| Books | 28.3% | 6.3% | 6.2% | 2.5% |
| News | 52.6% | 1.9% | 9.1% | 7.6% |
| Weather | 35.9% | 1.5% | 6.6% | 5.5% |
| Navigation | 46.3% | 4.6% | 5.9% | 3.3% |
| Travel | 41.1% | 4.4% | 6.3% | 7.1% |
| Shopping | 41.4% | 7.8% | 3.7% | 2.7% |
| Food & Drink | 38.4% | 8.3% | 2.6% | 4.4% |
| Medical | 35.5% | 4.1% | 3.9% | 7.9% |
Games complain about ads, almost never about billing. One game review in five is about ads. Surprise charges are at 0.8%, a tenth of what Food & Drink, Shopping, Health, and Education see. Games mostly sell through Apple's purchase sheet, where the price is on the button. The categories at the top of the billing column sell subscriptions, often with a trial in front.
News, Medical, Travel, and Weather readers blame updates. These are apps people open often for one quick task, so a moved button gets noticed.
News and Social Networking are the only categories where more than half of the reviews are negative. Books and Reference are the calmest, at 28% and 30%.
What a bad update looks like in the review feed
Because every review carries the app version it was written on, update blame can be pinned to a release:
| App | Version | First review | Reviews | Blame the update | Avg stars |
|---|---|---|---|---|---|
| Baby Tracker - Newborn Log | 5.0 | Jan 10, 2026 | 143 | 83% | 4.54 → 2.09 |
| Howbout: shared calendar | 16.3.0 | Sep 3, 2026 | 148 | 78% | 2.85 → 1.53 |
| Storm Radar | 4.0.12 | Apr 28, 2026 | 110 | 54% | 3.56 → 2.56 |
| MarineTraffic | 5.3.1–5.8.3 | Apr 27, 2026 | 456 | 70% | 1.1–1.7 |
| MyFitnessPal | 26.32.0 | Aug 11, 2026 | 143 | 52% | 2.00 |
| Opal: Screen Time Control | 4.7–4.8 | Jul 23, 2026 | 166 | 52% | 2.4–2.5 |
Baby Tracker is the cleanest case. Version 4.42.8 averaged 4.54 stars across 59 reviews. Reviews of version 5.0 and its new look start on January 10, and over the next nine days 118 of 143 blamed it: "We have been using the app for over 6 years with multiple children. It's a huge help - but the recent update made the design and usability much worse. Please revert back to the old visual design. Text is very hard to read." By May the app was back above four stars. Version 6.00 arrived on September 2, and 11 of its first 16 reviews blame the update again.
Howbout's version 16.3.0 turned event colours into "categories," five free and the rest paid. The feed turned over within days: "Howbout changed the color coding of events and took away a bunch of colors and put them behind a pay wall … I'm super annoyed by this and am now planning on switching to Google Calendar." A redesign and a new paywall arrived in the same release.
Storm Radar shows that recovery is possible. Its rebuild in April ("It was my #1 weather app since it's radar first and easy UI … Example: I can't find the humidity") dropped it a full star. Update blame peaked at 58% on 4.0.16. By 4.0.19 it is down to 11% and the average is back to 3.2.
MarineTraffic hasn't recovered. Its new look shows up in reviews from late April, and four months and eight versions later they still read: "I used to rely on this app on a daily basis. However, since the update I have switched to one of the competitors."
Eleven apps where most reviews describe a surprise charge
4.6% of all reviews describe a charge the author didn't expect, a subscription they couldn't cancel, or a refund they were refused. In most apps it's a handful. In 677 apps it passes 8% of what we analysed. At the top of the list it is most of them:
| App | Recent reviews describing it |
|---|---|
| Breeze: Self-Discovery Buddy | 420 of 500 (84%) |
| Novelmates | 416 of 500 (83%) |
| iDrama | 413 of 500 (83%) |
| Hypnozio: Sleep & Hypnotherapy | 335 of 458 (73%) |
| Sutara: Buddhist Stories | 221 of 310 (71%) |
| Liven: AI Life Coach | 309 of 500 (62%) |
| Ayatique: Quran Stories | 186 of 312 (60%) |
| Eatr: Tasty Cooking Recipes | 296 of 500 (59%) |
| Fabulous: Daily Habit Tracker | 272 of 500 (54%) |
| Offender Locator | 212 of 397 (53%) |
| GeoZilla | 260 of 500 (52%) |
These are claims made by reviewers. We haven't verified any of them, we haven't tested these apps' checkout flows ourselves, and a high share here says nothing about whether a company broke any rule. We did check the labelling by hand: of 60 flagged reviews drawn at random from these apps, all 60 describe a billing problem. Across random apps the flag is right about nine times in ten.
Reading them, reviewers describe two things again and again.
Reviewers say they paid on a website, and then couldn't find the subscription on their iPhone. In their accounts, the app or an ad led to a web checkout, the charge went to a card, and nothing appeared under Settings → Subscriptions, which is where every iPhone user has been taught to cancel. From an iDrama review: "It says to cancel through the App Store but my purchase was made through the app opening a website, not the AppStore." From Offender Locator: "The subscription also does NOT show up under my Apple Subscriptions." From Liven: "They want you to pay outside apple store so you get to follow their rules." iDrama also appears in our vertical drama study, where cancellation friction was already the loudest theme in that category.
Reviewers say they agreed to $1 and were charged more. From Breeze reviews: "I only consented to a $1 charge. That's all I authorized on Apple Pay. The screen clearly said $1." And: "I signed up for a $1 trial period of 7 days and they immediately charged my card for $19.99." From Eatr: "I only wanted to try the 1-week trial, but it automatically subscribed me to a 4-week plan and charged me $49.99."
If you work on one of these apps and think a number here is wrong, write to [email protected] and we will recheck it and publish a correction.
How we did it, and what it cost
Until now our review analysis used a general sentiment API. It returns a score per review: positive or negative, and how strongly. That tells you a review is angry. It doesn't tell you whether the author is angry about a crash or a charge, or whether they're leaving.
Jev works differently. You send text plus a list of questions, and each question has a typed answer: a choice from a list, a score against a rubric, or a probability between 0 and 1. We asked every review the same set:
- sentiment, and how confident the answer is
- kind of review (praise, bug report, design complaint, billing issue, feature request, other)
- severity of the problem, 0–3, and how much concrete information the review contains, 0–3
- three yes/no probabilities: surprise charge or blocked cancellation, author is leaving, author blames an update
- twelve topic probabilities
- language
Answers come back as numbers, so the whole dataset is one table we can query. Reviews in ten languages besides English, including Japanese, Russian, and Arabic, went through the same English questions with no translation step, and confidence stayed within 0.87–0.93 for every language.
The cost: Jev charges $42 per billion input tokens and nothing for output. A review plus our questions averages 1,453 tokens, almost all of it the questions. 1,074,607 reviews used 1.56 billion tokens: $65.60, or about $61 per million reviews. Another 89,160 reviews were too short to be worth a call ("Good", "love it") and were labelled from their stars for free. The run took about 25 hours.
This post isn't sponsored. We pay list price.
Where it's wrong
- 26% of negative reviews matched none of our twelve topics. The topic list has holes: games have no "gameplay" topic, and there's nothing for "missing feature" or "slow." Topic counts above are floors.
- 7.4% of answers came back with confidence under 0.6, mostly on mixed and neutral reviews.
- About one in ten reviews flagged as "five stars, negative text" is short praise the model misread.
- On random apps roughly 10% of surprise-charge flags are wrong. The misses are price complaints with no actual charge.
If you ship an iOS app
- Read your four-star reviews first. They come from users who like you and they contain a complaint about 70% of the time.
- Your paywall will generate more review text than your bugs. Plan the paywall copy and the cancellation path with the same care as the crash rate.
- Answer support email. It's the complaint most likely to end with "I've switched."
- Ship redesigns with a way back, or at least watch the first 48 hours of reviews by version. Baby Tracker's complaints started the day version 5.0 appeared.
- If you sell subscriptions on the web, say so in the app, and put a cancel button there. Reviewers can't find it in Settings, and they say so in public.
AppSigma pulls new reviews for the apps you track every day, each with its version, country, and developer reply, and serves them in the dashboard, the API, and the MCP server.
Method notes: 5,486 iOS apps from our tracked set, largest first; for each, the newest 500 reviews with text written in the 365 days before September 19, 2026, all storefronts. This is a sample of large apps' recent written reviews, not of the App Store: written reviews skew more negative than star ratings, and shares here should not be read as store-wide rates. 1,074,607 reviews were analysed by Jev (jev-latest); 89,160 with fewer than 15 letters were labelled from their star rating and are excluded from every percentage above except the total. A yes/no signal counts when its probability is 0.7 or higher. The stars comparison covers every tenth app: 107,501 reviews. App-level counts use all analysed reviews for that app. "Blames the update" by version requires at least 15 reviews on that version. Quotes are verbatim from public App Store reviews; charges, cancellations, and refunds described in them are the reviewers' accounts and were not verified.
Frequently asked questions
What do people complain about most in App Store reviews?
Price. In 1.07 million recent reviews of large iOS apps, pricing and paywalls came up in 195,188, more than all bug reports combined (186,464), and in 35% of all negative reviews. Ads were second overall and first in games, where one review in five is about them.
Do star ratings match what App Store reviews say?
At the ends, yes: 94.5% of one-star reviews have negative text and 81.5% of five-star reviews have positive text. In the middle, no. Praise with no complaint makes up 26% of four-star reviews, nearly half of three-star reviews are flatly negative, and roughly one five-star review in twenty reads as a complaint.
What does AI analysis of App Store reviews cost?
We paid $65.60 to analyse 1,074,607 reviews with Jev, about $61 per million. Each review was asked the same set of questions covering sentiment, review type, severity, twelve topics, and three yes/no signals. Almost all the cost is the question text, so cost per review is the same for short and long reviews and in every language.
Why can't I cancel a subscription in iPhone Settings?
Usually because the subscription was bought through a web checkout instead of Apple's in-app purchase. A subscription bought that way is billed to your card directly and doesn't appear under Settings → Subscriptions. Reviewers of several apps in this study describe exactly that. The subscription has to be cancelled with the company, or through the card issuer if the company doesn't respond.