Key takeaways
- Chat answers reflect the code the model learned from, so dependency versions and APIs in a chat-built project are often a generation behind.
- Make the build reproducible first: committed Gradle wrapper, one version catalog, a pinned JDK and a clean build from a fresh checkout.
- Test the minified release build before uploading – it behaves differently from the debug build you ran in Android Studio.
- Back up your upload key outside the project; with Play App Signing, a lost upload key can be reset, but it costs time.
What a ChatGPT-assembled Android project usually looks like
Most chat-built Android apps start the same way: a new project from an Android Studio template, then file after file pasted from ChatGPT conversations. It runs on the emulator, which feels like being almost done. Look closer and you typically find:
- dependency versions copied from answers, some current, some years old,
- Gradle snippets in Groovy pasted into Kotlin DSL files, or the other way round,
- XML layouts and Jetpack Compose screens side by side, without a plan,
- APIs that were standard in older tutorials and are deprecated now,
- an API key in a Kotlin constant or in
BuildConfig, - a debug keystore and no idea how release signing works.
The fix is not to start over. It is to turn the project into something a build server – and a coding agent – can understand.
Step 1: Make the Gradle build reproducible
A Play release has to be rebuilt months from now, on another machine, with the same result. Get there in this order:
- Commit the project to a private Git repository, including the Gradle wrapper files.
- Move all dependency versions into one version catalog (
gradle/libs.versions.toml) so there is exactly one place to read and update them. - Align the Android Gradle Plugin, Kotlin and Compose compiler versions – mismatches here cause most cryptic build errors in pasted projects.
- Pin the JDK version the build expects.
- Set
compileSdkandtargetSdkto 36. Google Play has required API level 36 for new apps and for updates since August 31, 2026, as our guide to the target API rule explains. - Prove it: a clean checkout builds with
./gradlew assembleDebugon a Linux CI runner.
This is ideal work for a coding agent that reads the whole repository and runs Gradle – the Codex CLI (included with paid ChatGPT plans), Claude Code or Gemini in Android Studio. Ask for one upgrade per pull request, with the build and tests green each time.
Talk to an AI Code Engineer about shipping your app to Google Play.
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.
Step 2: Replace what the chat learned from old code
Once the build is stable, ask the agent to list outdated patterns and replace them one area at a time:
| Found in chat-built code | Problem today | Replace with |
|---|---|---|
Overriding onBackPressed() | Deprecated; ignores predictive back, which apps targeting API level 36 get | OnBackPressedDispatcher callbacks |
AsyncTask | Deprecated, leaks and crashes on rotation | Kotlin coroutines with lifecycle-aware scopes |
| Kotlin synthetic view access | The plugin was removed from Kotlin | View Binding or Compose |
startActivityForResult | Deprecated | Activity Result APIs |
| Broad storage permissions for picking images | Policy scrutiny and a scary prompt | The system photo picker |
| All permissions requested at launch | Users deny, reviewers frown | Ask in context, when the feature needs it |
http:// endpoints | Cleartext traffic is blocked by default | HTTPS |
Also check edge-to-edge layouts: apps targeting recent API levels draw behind the system bars, and chat-written screens rarely handle insets.
Step 3: The release build is a different app
Debug builds are what you and ChatGPT ever tested. Release builds enable R8, which shrinks, optimises and renames code. Anything that relies on reflection – Gson mapping JSON to data classes is the classic case – can break silently: fields come back empty, or the app crashes on start.
- Build the release variant locally or in CI and install it on a device.
- Add keep rules or serialisation annotations for model classes, or switch to a reflection-free library such as kotlinx.serialization.
- Upload to the internal testing track and read the pre-launch report, in which Google lists crashes and warnings from test devices.
Step 4: Keystores, upload keys and API keys
New apps on Google Play are Android App Bundles signed through Play App Signing. Google keeps the app signing key; you sign uploads with your upload key. Generate that key once, store it with its passwords in a password manager, keep a backup and never commit it. CI gets an encoded copy as a secret, and the Gradle signing config reads it from environment variables. If the upload key is lost, Play Console offers an upload key reset, but it takes time you would rather spend elsewhere.
API keys are a different matter. A key in BuildConfig or a Kotlin constant ships inside the App Bundle and can be extracted. If your app calls the OpenAI API or another paid service, route the call through your own server, limit requests per user, remove the key from the Git history and rotate it. More in our playbook on secrets and API keys in AI-built apps.
Step 5: Through Play Console
- Account and testing. A new personal developer account needs twelve or more testers who stay in a closed test for fourteen days in a row before it can ask for production access – see our closed testing guide.
- Data safety. Declare what the app collects, including text and images your users send to an AI provider.
- AI-generated content. If the app generates text or images with AI, give users an in-app way to report offensive results.
- Accounts. In-app account deletion plus a web link for deletion requests – see the account deletion requirement.
- Payments. Selling premium features or AI credits inside the app means Google Play Billing.
Once the app is live, keep the agent on maintenance duty: yearly target API updates, library upgrades and fixes from Android vitals. If you would rather hand off the release work, our AI Code Engineers set up signing, CI/CD, testing tracks and Play Console for you, inside your own developer account, through our Android submission service. For the iPhone side, read app built with ChatGPT to the App Store.
Checklist: ChatGPT-built Android app before release
- The project builds from a clean checkout with the committed Gradle wrapper and one version catalog.
- compileSdk and targetSdk are 36; edge-to-edge and back navigation work.
- Deprecated APIs from chat answers are replaced.
- The minified release build was installed and tested on a device.
- The upload key is backed up outside the repository; CI signs releases.
- No paid API key ships in the app; leaked keys are rotated.
- Play Console items are done: closed test, Data safety, a report option for AI output, deletion link, Play Billing.
Frequently asked questions
Can I publish an Android app whose code came from ChatGPT?
Yes. Google Play reviews the app and your declarations, not the origin of the code. The app must meet the same technical and policy requirements as any other.
Why does my app work in Android Studio but crash after uploading?
The version from Play is a minified release build. R8 can break code that relies on reflection, such as JSON mapping with Gson. Build and test the release variant before you upload.
What happens if I lose my upload key?
With Play App Signing, you can request an upload key reset in Play Console. Until it is processed you cannot ship updates, so keep the key backed up.
Can you help us publish our app?
Yes. In a free call, one of our AI Code Engineers goes through your project and your Play Console status with you. If it helps, we then handle signing, CI/CD and releases on your behalf.
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.