The apps you can't leaveare the ones you rate worst

Read the chart left to right and it sorts itself into two kinds of app. On the left, Entertainment, Communication, Navigation, Social Networking, Utilities: a quarter of each is rated under four. These are the apps that come with a phone, a job, a country or a group chat. On the right, Games at 5%, Weather, Health & Fitness, Music, Education: the apps a person went looking for.

Nobody installs a messenger because they like it. They install it because everyone they know is on it. The rating is the one thing about that arrangement they get to decide, and so it carries everything: the missing feature, the forced update, the fee, the geopolitics. The rest of this article is about what happens when the only exit is a star.

On 15 November 2022, Ticketmaster's site fell over during the Verified Fan presale for Taylor Swift's Eras Tour and the general sale was cancelled. The company blamed bots. The reviews on Google Play blamed the app, and they have not stopped: a read of the 300 most recent this summer found 84% negative, with one in five about being flagged as a bot and one in seven about a barcode that would not show at the venue door. Then the record caught up with the reviews. A Senate committee read 110,000 pages of internal documents and found the Eras crash attributed internally to old legacy pages, not bots. The FTC sued over drip pricing that adds 24% to 44% at checkout. On 15 April 2026 a jury in New York found Live Nation and Ticketmaster liable on every antitrust count.

The Play Store rating had said all of that for three years, in the only language it has. This is the pattern for every app in this chapter. WeChat is the messenger a country runs on and its Google Play rating, written by the diaspora, is 2.9. Aadhaar is India's identity app: 6 million installs a month at 3.05, with reviews about a one-time password at every step. FIFA's own World Cup app went to 10 million installs a month in June at 3.2 stars and three in four recent reviews negative. Nobody in these reviews is choosing. They are reporting.

Google Maps is rated 4.7 on the App Store and 3.2 on Google Play. Ticketmaster 4.8 and 2.5. bilibili 4.8 and 2.0. Character AI 4.3 and 2.2. In our sample the median app rates 0.19 stars higher on the App Store, and 20% gap by half a star or more.

The mechanism has a number now. AppFollow read 51 million reviews this year and found that App Store star ratings sit about 0.6 stars above what the written reviews say, while Google Play stars sit 0.5 below theirs. The reason is who gets asked. On iOS the app chooses the moment, three times a year at most, with a silent one-tap prompt and no text box; the happy user taps five and moves on. On Android the review is written by someone who went to the store page, which is where you go when something is wrong. A game featured on the App Store averages 4.4 stars and 2.95 in its written reviews.

So the gap is partly instrumentation, and that is worth knowing before you copy a competitor's iOS number into a deck. But it does not explain Ticketmaster's 59% one-star, and it does not explain why the apps with the widest gaps are the assigned ones: the ticket seller, the campus messenger, the food chain's own ordering app. Android users are the ones who write it down.

Uber Eats is rated 4.8 on the App Store and 56% of its reviews in the last 90 days are negative. Amazon Shopping is 4.8 and 66%. Bumble 4.3 and 74%. McDonald's is 3.0 and two in three.

Here is the strange part. The American Customer Satisfaction Index surveyed 16,000 people this year and found satisfaction with food delivery going up: Uber Eats 76 out of 100, its best score yet. The same customers, asked in a survey, are content. Asked in a store review, they are furious. Both are true, because a store review is not a satisfaction survey. It is a complaint desk that happens to have a star on it, and the app is the desk. The pizza is late, the parcel is lost, the match was a bot, the seat did not exist, and the review lands on the software because the software is the only part of the service with a review button.

The same logic runs the other way. A star rating is a lifetime average. A bad quarter, a bad update, a bad price change hides inside it for years. The next chapter is about an app that found out.

CapCut grew on one promise: a serious editor, free, no watermark. Thirty-four million installs a month still arrive on that promise. Then the bill came. During 2025 the monthly plan went from $9.99 to $19.99 in many regions; in February this year the annual plan went from about $77 to $179.99, and existing subscribers rolled onto the new price on the 20th. Features that had been free moved behind the gate: a fade, a slow-motion, an audio extraction, a 1080p export. A read of 2,900 recent negative reviews found 31% about the paywall, and 70 people describing the same moment: an hour of editing, then an export button that asks for money.

The Google Play rating is 3.57 from 13 million ratings, 3.5 million of them one star. On the App Store it is 4.6. And the ads above are still selling the first promise. That is the whole mechanism in one app: the rating is where the gap between the ad and the product gets written down, one user at a time, and it is written by the people who cannot easily leave, because their projects are inside.

Gamers have the reputation for savage reviews, and the reviews are savage: AppFollow's benchmark has iOS gamers giving 4.3 stars while writing 3.6-star text. Yet Games has the happiest rating in the sample, 4.62 at the median, with only one game in twenty under four. The reason is not that games are better made. It is that a game that disappoints gets deleted, not rated. There is no school, employer or group chat keeping it on the phone. The unhappy categories are the ones where deleting is not an option.

Which means the rating is telling founders something other than quality. It is telling them where the audience is captive, and a captive audience with a low rating is a market with a door open.

A PDF reader at 3.55 with 6 million installs a month. A QR scanner at 3.64 with 4 million. Move to iOS, Apple's own switching tool, at 3.62: a one-use app rated by people at the exact moment it failed, so its ratings are 53% five-star and 44% one-star with almost nothing in between. Moovit, owned by Intel and installed in 3,500 cities, at 3.85. These are not hard products. They are big audiences being served badly, and the audience is counted on the store page for anyone to read.

It has been done, more than once. Niagara Launcher is six people, ten million downloads, no ads, no investors, and a 4.7 on Play in a category the phone makers own. Transit, an independent app, takes the daily-commuter reviews in cities where Moovit has the users. And Cal AI: two nineteen-year-olds and a team of seven went after a paywalled feature in MyFitnessPal, reached 15 million downloads, and in March MyFitnessPal bought them. The incumbent's rating was the map.

Start from a category, not an idea. Open Utilities, Navigation, Social or Communication, sort by installs, and stop at the first app under 4.0. In our sample that is one app in four in each of those categories, and the biggest of them have millions of installs a month.

Read the Play reviews, not the iOS ones. The App Store number is inflated by the silent prompt and the Play number is where people write down what is wrong. The last hundred reviews of a low-rated incumbent are a product spec someone else paid to research.

Build the thing the top complaint describes, and price it honestly. CapCut's 3.5 million one-star ratings are mostly about a promise that changed. The gap between an incumbent's ad and its product is the room a small team gets to stand in.