Google Play Store Listing Experiments compare versions of your listing using a metric selected in Play Console. Start with a specific question about the icon, screenshots or wording. A test can favor the current listing, favor a variant, produce a draw or need more data; running one does not guarantee an improvement.
This guide follows the current Google setup and result documentation, checked September 11, 2026. The console's eligibility and displayed definitions should govern your actual experiment.
What can you test?
Default graphics experiments cover the icon, feature graphic and screenshots. Localized experiments also support descriptions. Google currently permits one default graphics experiment or up to five localized experiments for an app, with up to two experimental variants against the current listing.
Language eligibility matters: a default graphics experiment does not necessarily include everyone who visits a multilingual listing. Check the audience shown in the console. These experiments do not compare arbitrary app versions or reveal Google's search-ranking weights.
For an Apple app as well, use the separate cross-platform experiment guide. Apple and Google have different supported assets and metrics.
Choose a hypothesis worth testing
Name the visitor problem before deciding which asset to change. Examples: "People cannot tell that our workout app works offline" or "The first screenshot explains setup before showing a completed result." A meaningful alternative should change how the app is understood.
| Asset | Possible hypothesis | Keep stable where practical |
|---|---|---|
| Icon | A simpler shape makes the app easier to recognize | Other listing assets and campaign message |
| First screenshot | Showing the finished task explains the benefit sooner | The rest of the gallery and metadata |
| Feature graphic | A readable task-focused composition works at small size | Other graphics and product claim |
| Description | A specific supported use case reduces ambiguity | Scope of the product and audience |
These are hypotheses, not findings. Do not copy a competitor's visual simply because the app is popular; its audience, brand and acquisition mix may differ.
Prepare the current version and one alternative
Save the current asset before editing. Change one asset where practical so the result is easier to interpret. If you intentionally test a whole campaign, report it as a campaign comparison rather than attributing the outcome to a single color or caption.
Use the screenshot editor to prepare a consistent pair of designs. Manual editing and export of Free templates require a free account; Pro collections have separate access requirements and AI work may consume tokens.
Inspect the exported images at native resolution and at the size a visitor sees. Follow the store's content rules, including the distinction between mandatory requirements and recommendations. A feature graphic, phone screenshot and Wear OS image do not have identical presentation rules.
Select the metric before launch
The current console offers unique-user install, open and pre-registration click metrics. Choose the one appropriate to your audience and goal, and save its displayed definition in the experiment record. Do not relabel the result as paid customers or retained subscribers.
A description experiment evaluates the selected acquisition metric; it does not establish how the wording affects organic keyword rank. If you also maintain keyword observations, report them separately.
For a subscription product, inspect later payment and renewal cohorts as a separate outcome. A more persuasive listing can increase installs while producing poorly matched users who never complete the task.
Set up the experiment in Play Console
- Open the app's Store listings area and choose the available experiment setup for the listing.
- Name the test so the hypothesis is recognizable later.
- Choose the default graphics or localized experiment type and the target metric.
- Review variants, audience allocation, smallest change the test is designed to detect and confidence settings.
- Upload the alternate asset and inspect the previews.
- Save and follow the publishing review flow before assuming the experiment is active.
Confirm that the experiment is running and record its start date. Keep a screenshot or export of the settings so a future reviewer can understand the eligible population and method.
Decide whether the expected duration is practical
Use the console's estimated duration and sample requirement. Smaller effects, more variants and less eligible traffic can require more evidence. A low-volume app may learn more from repairing an obvious error or testing a larger, meaningful difference than from testing a subtle color change.
Let the required sample guide how long you run the test. Include an appropriate mix of weekdays and weekends where it matters, but time alone does not establish a result. Google currently stops experiments after six months without applying a variant.
Interpret the result as reported
Read the explanation beside the result, including the confidence interval and smallest change the test is designed to detect. "More data needed" is not a failure and is not permission to declare the current leader a winner. A draw may mean the difference is smaller than the chosen decision threshold.
Keep the estimated difference, uncertainty, audience and dates together. If the current listing performs best, keeping it is a valid decision. If more than one variant performs well, consider which communicates the product most accurately and is easiest to maintain.
For an illustrative calculation, moving from 5% to 7% acquisition conversion is a two-percentage-point increase and a 40% relative increase. That arithmetic does not establish statistical confidence or predict your test result.
What can make a result hard to interpret?
- Stopping on an early lead: a favorable estimate may change as data arrives.
- Changing several things: the result cannot identify which individual change mattered.
- Combining audiences: different languages and acquisition sources may respond differently.
- Ignoring eligibility: the experiment may exclude localized traffic you assumed was included.
- Calling observation causation: a concurrent rank change is not automatically caused by the tested asset.
- Ignoring downstream outcomes: more installers can still be less likely to complete useful work.
Record the decision and prepare the next useful test
Use an experiment log with the question, original version, alternative, settings, dates, reported result and decision. Preserve negative and inconclusive outcomes. Use what you learn to choose the next test; some changes will not help.
The ASO report template provides a place to connect the test with other listing work. If the main problem occurs after installation, work on that next rather than testing continuously for its own sake.
Frequently asked questions
How many variants can I test on Google Play?
The Google documentation checked September 11, 2026 allows up to two experimental variants against the current listing. Confirm the options and eligibility shown in your Play Console.
Must I wait for 95% confidence?
Use the confidence setting, interval, smallest change the test is designed to detect and result explanation for your experiment. A fixed percentage rule should not replace the method configured in the console.
Can a listing experiment prove a keyword ranking improvement?
No. It evaluates the selected store experiment metric. Keyword ranking observations are separate and may change because of factors outside the tested listing asset.
