Key takeaways
- The closed test checks real engagement and whether you acted on feedback, not just the opt-in count.
- An agent's real value is cadence: when a fixed build ships in minutes, you can respond to feedback throughout the 14 days.
- Automate uploads, release notes and the feedback log; never automate testers, engagement or invented feedback.
- Draft the production access answers from the log only, then edit them yourself.
- Recruit 15 to 20 real testers so one dropout does not restart the clock.
The rule in one paragraph – and where the details are
If your Play developer account is a personal account created after November 13, 2023, you cannot publish to production until a closed test has run with at least 12 testers who stayed opted in for 14 days in a row. Then you apply for production access and answer questions about the test, the app and its readiness. Organization accounts are not covered by this requirement. Google's help page on testing requirements is the official source, and our closed testing guide covers eligibility, tester setup, rejection reasons and the application form in depth. This playbook is about running the test with a coding agent in the loop.
Why AI-built apps struggle with the closed test
Google's decision looks at whether testers really used the app and whether you acted on what they reported. Apps built quickly with agents often enter the test with one build and no plan: testers install, open the app once and forget it, and two weeks later there is nothing to report beyond "testers liked it". The opposite shortcut is just as risky – services that sell opt-ins without real use. Both end the same way: a request for more testing and another 14 days.
This is where an agent genuinely helps. When turning a bug report into a fixed, uploaded build takes minutes instead of an afternoon, you can respond to every useful report during the test. Those responses – visible as new builds with release notes – are exactly the evidence the application asks for.
Talk to an AI Code Engineer about your Google Play closed test.
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 three-week plan
| When | Human | Agent |
|---|---|---|
| Before day 0 | Recruit 15–20 testers who resemble your users; set up a Google Group or email list; write the onboarding message | Build the upload lane for the closed track; fix crashes from the first pre-launch report; set up the feedback log |
| Day 0 | Publish the first closed release and send the opt-in link | Record each tester's opt-in date in the log |
| Days 1–7 | Send a weekly task list; watch the opted-in count | Group incoming feedback into issues; ship fixes; draft release notes per build |
| Days 8–14 | Send the second task list; confirm everyone is still opted in | Ship further builds; summarise feedback themes weekly |
| Day 15 onward | Apply for production access with answers based on the log | Draft answer text from the log for you to edit |
| While Google reviews | Keep the test running and testers engaged | Keep fixing and shipping |
Google states that the application review usually takes seven days or less but can take longer, so the whole exercise is closer to three weeks than two. Plan your launch date accordingly.
What to automate
- Track uploads. Every merged fix becomes a closed-test build with a higher versionCode and release notes naming what changed. In the Play Developer API, the default closed track is addressed as alpha; fastlane's supply and Gradle Play Publisher use the same track names. Our fastlane playbook shows a matching lane.
- Release notes. Drafted from merged pull requests, then shortened by a person into language testers understand.
- Feedback log. Reports from email, the tester chat and Play's private testing feedback gathered into one table with date, tester, issue, action and build. The agent keeps it current; you check it.
- Pre-launch report triage. Each crash or ANR that Google's test devices find becomes an issue with device and Android version.
- Policy pre-checks. Target API level, permissions, Data safety alignment and reviewer credentials, using the Google Play review checklist.
A closed test is also the first time the release build meets many different devices. Problems that only appear on one manufacturer's phones or an older Android version surface here – an advantage, as long as someone turns them into fixes.
What must never be automated
When an agent can script almost anything, some shortcuts look tempting. Do not take them:
- Fake testers or scripted activity. Emulators or throwaway accounts that open the app every day misrepresent your test to Google and can put the developer account at risk – a far bigger problem than a rejected application.
- Invented feedback. An agent asked to "write a strong summary of tester feedback" produces one whether or not the feedback exists. Give it the log and instruct it to use nothing else.
- Shared credentials. Never give a tester service, or an agent that can reach the internet freely, your Play Console login.
Google's User Ratings, Reviews, and Installs policy forbids inflating installs by illegitimate means. The testing requirement exists to see real use; give Google real use to see. If a tester leaves, recruit a replacement early – their 14 days start when they opt in.
Writing the application with evidence
The production access form asks how testers engaged, what they reported and what you changed. With a complete log, the agent can draft answers that cite real numbers: how many testers finished the core flow, which issues came up, which builds fixed them. A prompt that keeps it honest:
Draft answers for the Google Play production access application using ONLY the entries in feedback-log.csv and the release notes in /releases. Quote numbers exactly as they appear in those files. If the files do not support a statement, leave it out and list it under MISSING. Keep each answer under 150 words.
Then edit the text yourself – you are the one submitting it, and you know things the log does not. After approval, every production release still goes through Play review, so keep the same cadence. TestFlight is the iOS counterpart of this loop, without a mandatory tester count; see TestFlight for AI-built apps.
Our AI Code Engineers help plan the test, build the pipeline that ships builds to the closed track and review your answers before you apply – part of the Android submission service.
Closed testing checklist for agent-built apps
- You recruited 15 to 20 real testers who resemble your target users.
- Testers opted in to the closed track, not the internal track, and the app is available in their countries.
- Every merged fix reaches the closed track automatically with a higher versionCode and release notes.
- A feedback log records date, tester, issue, action and build for every report.
- At least two builds during the test address problems testers reported.
- Pre-launch report crashes and ANRs were fixed before you applied.
- Reviewer credentials, Data safety, content rating and target audience are complete.
- Application answers were drafted from the log only and edited by you.
- No scripted testers, bought opt-ins or invented feedback were used.
Frequently asked questions
Can an AI agent act as one of my 12 testers?
No. Testers have to be real people with their own Google accounts who use the app on their devices. Scripted activity misrepresents the test and puts your developer account at risk.
How many builds should we ship during the closed test?
Google sets no minimum, but it expects you to act on feedback. Two or more builds that fix problems testers reported, each with clear release notes, give you something concrete to describe in the application.
Our application was rejected. What now?
Continue or restart the closed test with engaged testers, ship fixes based on their feedback and apply again with more specific answers. Our closed testing guide lists the typical reasons and how to address each one.
Can you help us pass the closed test?
We cannot supply testers, and we would not want to. Book a free call and an AI Code Engineer helps you plan the test, automate the builds and prepare a well-documented application.
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.