Key takeaways
- Split the review into a code scan by the agent, a device test by a person and a metadata pass by the account holder.
- Ask the agent for evidence with file paths and line numbers, not for a yes-or-no verdict.
- AI-built apps often carry leftovers: placeholder screens, debug menus, staging endpoints and features nobody described to the reviewer.
- Apps that send personal data to a third-party AI service must say so clearly and ask for permission first (Guideline 5.1.2(i)).
- Each check links to a detailed guide on the matching guideline instead of repeating it here.
How to use this checklist with a coding agent
App Review does not care which tool produced your app; it cares about what a reviewer sees on a fresh install. The difficulty with agent-built apps is volume. Dozens of sessions add features, flags and dependencies faster than anyone keeps track of them, and the person submitting often no longer knows everything the app contains. So this checklist is split by who can verify each item reliably:
- Pass 1 – code scan. The agent searches the repository and reports findings with file references. It fixes nothing yet.
- Pass 2 – device test. A person installs the TestFlight build on real hardware and walks the reviewer's path.
- Pass 3 – App Store Connect. The account holder checks every field, answer and note against what passes 1 and 2 found.
Fixes come afterwards, one task each. Scanning and fixing in the same session tends to end in an "all good" summary that includes changes nobody tested.
Pass 1: the code scan your agent can run
| Check | What the agent should report | Background |
|---|---|---|
| Leftovers | Filler text, sample data, "coming soon" screens, dead buttons, to-do comments in UI code | 2.1 App Completeness |
| Debug paths | Debug menus, hidden gestures, staging or localhost URLs, test keys reachable in release builds | 2.1 and basic security |
| Hidden features | Feature flags, remote config switches, role-gated or paywalled screens, and how to reach each one | 2.3.1 and AI-built apps |
| Payments | Every path to Stripe, PayPal, web checkout URLs or "upgrade on our website" copy | 3.1.1 In-App Purchase |
| Login | Third-party sign-in providers and whether an equivalent option such as Sign in with Apple is offered | 4.8 Login Services |
| Account deletion | Where accounts are created and where in-app deletion is implemented, end to end | 5.1.1(v) |
| Permissions | Every usage description in Info.plist, the code that triggers it and whether the text names a concrete purpose | 5.1.1 |
| Tracking and SDKs | Ads, analytics and attribution SDKs, App Tracking Transparency use, privacy manifest coverage | 5.1.2 |
| Third-party AI | Calls to external AI services that receive personal data, and the consent step before them | 5.1.2(i) |
| Web content | Which screens are WebViews of an existing website | 4.2 Minimum Functionality |
Ask for the result as a table with a file path and line number per finding and an explicit note where the agent is unsure. Pass 3 uses this table again.
Talk to an AI Code Engineer about getting your AI-built app through App Review.
In a free call we look at what you built, tell you honestly what stands between it and an approved release – and whether you need us at all.
A prompt for the scan
Review this iOS project for App Store submission readiness. Do not edit files. Report each finding with file path and line number, or "none found": 1. Placeholder content, sample data, unfinished screens, debug menus. 2. Non-production URLs or keys reachable from release builds. 3. Features behind flags, roles, remote config or paywalls, and how to reach each. 4. Every payment path; flag anything that is not StoreKit for digital goods. 5. Sign-in providers and whether Sign in with Apple is offered next to them. 6. Account creation and the in-app account deletion flow. 7. Info.plist usage descriptions and the code that triggers each permission. 8. Third-party SDKs, tracking, and PrivacyInfo.xcprivacy coverage. 9. Personal data sent to external AI services, and where consent is asked. Mark anything uncertain as UNSURE instead of guessing.
Run the same prompt again on the exact commit you archive. A clean second report on the submitted code is worth more than the first one, because fixes made in between can introduce new leftovers. If your app comes from an AI app builder rather than a repository you control, export or sync the code first – the scan needs files to read.
Pass 2: the device test only a person can do
Agents test what they can run: compilers, unit tests, perhaps UI tests on a simulator. Reviewers use real devices, fresh installs and imperfect networks. Install the TestFlight build and work through this order:
- Remove any previous install, then launch without an account. Does the first screen make sense without context?
- Sign up as a new user, sign out, sign in again – including with Sign in with Apple if offered.
- Deny every permission prompt once. The app must keep working wherever it reasonably can.
- Use the core feature end to end with the demo account you will give the reviewer.
- Buy and restore in the sandbox if you sell anything, and check that price and terms are visible before the purchase.
- Delete the account in the app and confirm on the backend that it is gone.
- Repeat the main flow on an iPad if the app runs there; see Guideline 2.4.1.
- Try airplane mode and a slow connection on every screen that loads data.
Each problem becomes one agent task with steps to reproduce. Our TestFlight playbook describes how to run that loop with testers.
Pass 3: the App Store Connect fields only you can answer
- App Privacy. The answers match the SDK and data list from pass 1, including data that goes to AI services.
- Age rating. The questionnaire reflects everything the app can show, including AI-generated and user-generated content.
- Review notes. A demo account that works without one-time codes, plus a description of every feature from the hidden-features row of pass 1. Our post on notes for review has examples.
- Screenshots and description. Taken from the build under review, describing features that exist; see Guideline 2.3.
- Support and privacy policy URLs. Live pages with a working contact option; see Guideline 1.5.
- In-app purchases. Products attached to this version, with review screenshots and complete metadata.
Risks that are specific to AI-built apps
Three patterns show up more often in agent-built apps than in hand-built ones.
Generic design. Builders and agents converge on the same layouts, icons and onboarding copy. An app that looks like many others and mostly wraps a website invites questions under Guideline 4.2 or 4.3. Give it a clear purpose and native functionality worth installing.
Behaviour that changes after review. Remote config, server-driven screens or a WebView that loads new features mean the reviewer approved something other than what users get. Our post on Guideline 5.6 and AI-built apps explains why that is risky.
AI features that share personal data. Since November 2025, Guideline 5.1.2(i) explicitly requires apps to disclose when personal data goes to third parties, including third-party AI, and to obtain permission before sharing it. A chat or photo feature that sends user content to an AI API needs a visible consent step, not only a line in the privacy policy.
If you want a second pair of eyes before submitting, our AI Code Engineers run this review on your build and can take over the submission in your own account – see the iOS submission service. The Android counterpart of this list is the Google Play review checklist for AI-built apps.
Final App Store pre-submission checklist
- The agent's scan report on the archived commit shows no placeholders, debug paths or staging URLs.
- Every hidden, flagged or paywalled feature is listed with instructions in the review notes.
- Digital goods and subscriptions use In-App Purchase; any external links follow the current rules for each storefront.
- Third-party login is paired with an equivalent privacy-friendly option such as Sign in with Apple.
- Accounts created in the app can be deleted in the app.
- Permission texts name a concrete purpose, and the privacy manifest covers the app and its SDKs.
- Personal data sent to AI services is disclosed in the app and only sent after the user agrees.
- A person completed install, sign-up, core feature, purchase, restore and deletion on iPhone and, if supported, iPad.
- App Privacy answers, age rating, screenshots and description match the build under review.
- Support and privacy policy URLs load and offer a way to contact you.
Frequently asked questions
Does Apple reject apps because they were built with AI?
Apple has no rule against apps built with AI tools. Reviewers apply the same guidelines to every app. AI-built apps simply tend to hit some of them more often – completeness, hidden features, login, payments and privacy – which is why this checklist focuses on those.
Can the agent run the whole checklist on its own?
Only pass 1. The device test needs a person on real hardware, and the App Store Connect answers are statements the account holder makes. The agent's report makes both of those passes faster and more accurate.
The agent's scan found nothing. Can I trust that?
Check how it searched before you trust the result. Ask which directories and file types it covered, whether it included dependencies and generated code, and spot-check two or three items yourself, for example by searching for your staging domain.
When should we run the checklist?
Run pass 1 before the release candidate goes to TestFlight, pass 2 on that TestFlight build, and pass 1 again on the exact commit you archive for submission. Pass 3 happens in App Store Connect right before you press Submit.
Can you check our app before we submit?
Yes. Book a free call and an AI Code Engineer goes through your build with you. If you want, we handle the submission and the review conversation in your own Apple Developer account.
Free, 15 minutes, no obligation
Get your app from agent output to approved release.
Our AI Code Engineers take apps built with Claude Code, Codex, Cursor, Lovable, Base44 and co. through signing, CI/CD, store listings and review – in your own developer accounts.
AI tools, store policies and requirements change quickly – always check the current documentation of the tools and stores you use. This article is independent advice and not affiliated with or endorsed by Apple, Google, Anthropic, OpenAI or any other company or product mentioned; all trademarks belong to their owners.