Dev.to WebDev ๐Ÿ›  Dev ๐Ÿ‘ 0 ๐Ÿ“– 4 min read

Five App Store rejections I keep seeing in AI-built apps, and the metadata that prevents them

A founder sent me a screen recording last month: the app worked, the payment worked, the icons were done. The rejection was three lines long and mentioned a guideline number. Two weeks had already gone by. The build itse

A founder sent me a screen recording last month: the app worked, the payment worked, the icons were done. The rejection was three lines long and mentioned a guideline number. Two weeks had already gone by. The build itself took four days. The review queue took longer than the code.

I build apps for a living โ€” solo, from Seoul, three to five days per project, usually finishing something that Cursor, Lovable, Bolt or v0 got to about 80%. So I read a lot of rejection notices. They repeat. Here are the five I see most often in AI-built apps, and the cheap fix for each.

1. Minimum functionality (4.2)

The app is a web view in a native shell and adds nothing the browser could not do. Reviewers are not hostile to wrappers โ€” they are hostile to wrappers with no reason to exist. A shell is accepted when it does something the browser cannot: offline play, push notifications, camera, local storage that survives, a real game loop.

What to prepare before submitting: one sentence in the review notes explaining what the native shell provides. If you cannot write that sentence honestly, the app is not ready and the guideline is right.

2. Crashes and placeholder screens (2.1)

AI tools generate a lot of screens. Some of them are empty states that were never finished. A reviewer who taps a button and lands on a blank page rejects the app, and they are correct to. This is the single most common reason a build that "works on my machine" fails: nobody walked the app twice, end to end, on a clean account.

Fix: create a fresh account and use the app for fifteen minutes without touching the code. Every dead end you find there is a rejection you just avoided.

3. Privacy strings and the missing delete button (5.1.1)

Two separate items, both in 5.1.1. First, every permission needs a purpose string that says what it is used for, in the user's language, not "needed for the app". Second, an app that creates accounts must offer account deletion from inside the app โ€” not a support email, not a web form. This one catches a lot of AI-generated apps that use a generic auth template, because the template ships sign-up and never ships deletion.

4. Digital goods paid outside the store (3.1.1)

If the app sells something that is consumed inside the app โ€” credits, a subscription to content, a feature unlock โ€” Apple expects in-app purchase. A Stripe link on a page for the same digital item is the classic rejection. Physical goods and services are different; a real-world booking or a shipped item can be paid outside.

Decide this before you build the checkout, because moving a payment flow later means touching the screen, the backend, the receipt validation and the store products at once.

5. Metadata that does not match the app (2.3)

Screenshots that show features not in the build, a description promising a platform that is not supported, an age rating that ignores user-generated content. Guideline 2.3.1 is short and unforgiving. It is also entirely in your control, which makes it the cheapest item on this list to fix.

Play Store has its own version of the same problem: the Data safety form and the target API level requirement. Neither is about code, and both can hold a release.

What this costs in days

  • Metadata and privacy text: half a day if you write it once, correctly.
  • A second review round: typically five to ten days of queue time you did not plan for.
  • Moving payments to in-app purchase: one to two days, plus a day of testing with sandbox accounts.

That is why I treat the store as part of the five days, not a step after them. A build that is finished but not accepted is not shipped.

How I price it

  • Before the contract: 0. The conversation and the estimate are free.
  • You pay nothing until you see working screens. The first payment comes during the work, the rest after you have reviewed the result.
  • Fixes inside the agreed scope are free until you are satisfied.
  • Contract and invoice: CPC AI Inc., a Korean company, registration number 771-88-03539, in English.
  • Typical price for three to five days: $900โ€“1,800, up to 1/20 of what an agency asks for the same scope on a three to six month schedule. The difference is not typing speed; it is that there is no sales team, project manager and subcontractor between you and the person writing the code.

Seoul is UTC+9. If you are in Europe or the Americas, written work moves overnight: you send the question in the evening, the answer is there in your morning.

Five lines are enough for an estimate

  • What the app does โ€” one sentence.
  • Who uses it โ€” a customer, an employee, or both.
  • What already exists โ€” a link, a repo, screenshots, or a spreadsheet.
  • Platform: web, iOS, Android, or all three.
  • A deadline that is real.

Write to [email protected] โ€” I answer in English, and I can read Korean, Japanese, and most European languages with a translator.

Work and examples: https://cpcai-studio.netlify.app/en/

The rejection notice is not a verdict on your idea. It is a list of things that were never written down. Write them down before the first submission and the five days stay five days.

๐Ÿ“ฐ Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes โ€” full credit and traffic to the original publisher.