Apple Rejected 2 Million App Submissions in 2025. Performance Outweighed Every Other Reason Combined

Apple's 2025 App Store Transparency Report has a number I keep coming back to. App Review looked at 9,100,620 app submissions in 2025 and rejected 2,093,244 of them. That is roughly 23%, a little under one in four.

The more interesting number is underneath. Apple breaks rejections down by the section of the App Review Guidelines they were cited under. Performance was cited 1,354,418 times. Legal, Design, Business, Safety and Other together add up to 1,350,329. The single biggest reason apps bounce off the App Store is a section about whether the app is finished and works, and it outweighs everything else on the list combined.

One caution before reading too much into that. Apple notes that a submission can be rejected under more than one guideline, so these are citations, not unique apps. It still tells you where the friction is.

What "Performance" actually means

When people hear Performance they think speed. In the App Review Guidelines, section 2 is broader and more boring than that. It covers App Completeness, Beta Testing, Accurate Metadata, Hardware Compatibility and Software Requirements.

App Completeness is the one that bites. Apple asks that submissions be final versions with all the metadata and working URLs, with placeholder text, empty websites and other temporary content scrubbed out. It says plainly that it will reject binaries that crash or show obvious technical problems.

Accurate Metadata asks that your description, screenshots, previews and privacy information reflect the app's core experience, and that you keep them current as the app changes.

Read those two together and you get a picture of what a lot of rejected apps look like. They were submitted before they were done, with screenshots that describe a version that no longer exists, a privacy answer that was filled in once and never revisited, a settings link that goes nowhere.

Why AI makes this number go up

I build and ship iPhone apps myself, mostly with AI doing a large share of the typing. I have written about how that works. What AI changed for me is the time between an idea and a build that runs. What it did not change is the time between a build that runs and an app that is finished.

Those are different stages, and only the first one got cheap. A model will happily generate a settings screen with a "Terms of Service" row that opens a URL it invented. It will write a paywall with a restore button that does nothing yet, because you did not ask for the restore flow. It will produce an onboarding screen with lorem ipsum in the third card because the first two were the ones you looked at. Each of those is a guideline 2.1 rejection waiting to happen, and none of them shows up when you run the app on your own phone and tap through the path you already know.

The report shows 9.1 million submissions in a year. I do not know how many of those were built with AI help, and Apple does not say. But I know which direction the cost of producing a submission went, and it is down. When producing something gets cheaper and checking it does not, the share of unfinished things that reach the reviewer goes up.

Design is the third biggest bucket

Design guidelines were cited 415,532 times. Apple's own note on the report says this category includes the rules against copycat apps, apps with minimum functionality, and spam.

Guideline 4.2 is the one I think about most. It says an app should include features, content and UI that elevate it beyond a repackaged website. That is a product question. A wrapper around a web view is easy to generate now. A reason for the app to exist on the phone, something it does better because it is installed, still has to come from someone who thought about it.

Copycats and spam fall in the same bucket for a reason. If a model can produce a plausible clone of a popular app in an afternoon, then the store fills with plausible clones, and the reviewers become the people who say no on behalf of everyone else. Apple also removed 166,899 apps from the store in 2025 according to the same report. The review is not the last gate.

The review is your first user test

Here is how I think about it now. App Review is the first time someone who has never seen your app opens it with no context and tries to use it. That is a usability test. It just happens to have a binary outcome.

So I run my own version before I submit, and it is mostly the things a reviewer will hit in the first two minutes:

A fresh install on a device that has never seen the app, signed out, with no test data.
Every link in settings and the footer tapped once, including terms, privacy and support.
Restore purchases, sign out, delete account, and the empty state after each one.
The screenshots and description read side by side with the build, looking for anything the app no longer does.
A search for placeholder text in the project, because a model will leave some, and so will I.
If the app needs an account, a demo login in the review notes that actually works.

None of this is clever. That is the point. The reasons in section 2 are mostly unglamorous, and the fix for unglamorous problems is a checklist you actually run.

Rejection is not the end

One more figure from the report: 387,087 submissions were approved after an earlier rejection. A rejection is a round trip, and the guidelines are specific enough that most of them are fixable in a day.

But a round trip still costs a launch date, and if you planned a release around a press moment or a store feature, a day can be a lot. The cheapest place to catch an incomplete app is before it leaves your machine.

The headline of this report is that Apple rejected two million submissions. I read it as two million times someone was faster at building than at finishing. AI is going to make that gap wider for everyone who lets it. The people who ship cleanly will be the ones who spend some of the time AI saved them on the boring last mile.