Skip to content
AppDues
Log in
Contact salesSign up

Google App Campaigns: Setup, Optimization, and Measurement

Google app campaigns can deliver scale with little day-to-day management, but that automation comes with a trade-off: fewer controls, modeled reporting, and a slow learning period. This guide explains what the channel is, which inputs actually matter, and how to avoid mistaking platform-reported activity for incremental growth.

What the channel is — and what you cannot control

Google app campaigns, also called GAC and formerly Universal App Campaigns, are the app-advertising product for Android and iOS. Think of them as an automated, app-specific counterpart to Performance Max: there is no manual campaign layer where you pick keywords, apps, channels, audiences, or devices.

Delivery can span Google Search on web and in apps, YouTube, Play Store search and display on Android, and the Google Display Network, mainly AdMob. That breadth is the main attraction. Google properties are not available through most other app channels, and the inventory can scale to very large volumes with minimal manual maintenance.

The three campaign types are app install, app engagement or retargeting, and pre-registration for apps not yet launched on Google Play. Most practical guidance focuses on install campaigns optimized toward in-app events.

Android versus iOS, and a launch checklist

Android is usually the stronger fit. Play Store placements are exclusive to Google, and Google can observe more of the journey, which supports optimization. iOS inventory is narrower — mainly AdMob display, Google Search, and YouTube — and, in the period covered here, Google did not optimize from SKAdNetwork data. Instead it modeled conversions from engagement and the limited events it could observe, because it does not request ATT permission in its own apps.

A useful launch checklist:

  • Confirm conversion tracking through Firebase or an MMP before spending, and link Play Console for Android.
  • Pick one primary optimization event per campaign; avoid running two campaigns with the same event in the same country.
  • Set budget and bid high enough to produce at least about 10 events per day, with 50 or more preferable, while remembering Google may spend above the daily budget early in learning.
  • Change location targeting from the default interest-based setting to people in your targeted locations, so ads do not drift into unselected places.
  • Bundle countries with similar value per event rather than spreading a thin budget across many small campaigns.

Choosing the event and bid

The setup wizard asks for five main inputs: daily budget, optimization goal, bid, location and language, and creatives. Goals range from installs, to installs likely to complete an in-app event, to in-app events, to value optimization when revenue is passed. The bid is a target cost per action, or tCPA, for event goals, and a target return on ad spend, or tROAS, expressed as a percentage, for value goals.

Install-only campaigns may look cheaper, but they often bring low-value users. For subscription apps, trial starts are commonly the best compromise: converted trials and renewals are usually too delayed and sparse to optimize against, while trials happen early enough and correlate reasonably with revenue. If plans differ greatly in value, bundling all subscription starts into one event is simpler than splitting them; value-based tROAS needs more data, higher budgets, and reliable per-plan revenue prediction, and was designed mainly for gaming and commerce.

If delivery stalls because the tCPA is too low — a situation called overshooting — consider raising the bid, moving to a higher-funnel or more frequent event, refreshing assets, bundling countries, or leaving the channel if it cannot deliver at an affordable price.

Building an asset set that can learn

Creative assets are the other major lever. Google recombines text, images, video, and HTML5 into ads, and reporting shows asset performance rather than the exact combinations users saw. Missing a format can mean missing a placement: without video, for example, YouTube delivery is limited.

Aim for at least two or three assets per category, up to the limit of 20 per category. Cover image ratios around 1.91:1, 6.4:1, and 1:1. For video, test portrait, square, and landscape at several lengths, including 6, 10, 15, and 30 seconds. Write headlines and descriptions so each one still makes sense when recombined, and test radically different messages before producing minor variants of a winner.

Ad groups can organize assets by theme, such as an evergreen set of winners plus rotating seasonal or message-specific sets. Judge assets within a category using the asset tab, but do not rely only on Google's Best, Good, or Low rating — check delivery, engagement rate, CTR, and modeled CPI and CPA as well. Replace persistent low performers with new concepts.

Learning, reporting, and the search blind spot

Machine learning needs time and volume. Expect a learning period measured in weeks and, as a rough guide, at least 100 events before judging a campaign. Frequent changes can reset learning, so raise budgets gradually — for example by about 20% once or twice a week — and prefer waiting and analyzing over constant edits.

For reporting, segment campaigns by network, device, or day in the interface, but note that day segmentation is limited to two weeks. API access or a third-party tool is better for longer trends and for joining cost and conversion data. Also check the hidden app campaign placements report for which apps showed your ads, while recognizing it now shows mainly impressions.

Search deserves special attention. You cannot select keywords, run search-only campaigns, or see keyword reporting, and branded searches can inflate apparent performance without being incremental. Play Console search-term data, account-level negative keywords where available, and controlled add-or-remove tests on brand terms can give partial signals. A text-only campaign is not a reliable way to force search-only delivery, because Google can generate ads from your store listing for its app network.

Testing incrementality before trusting the dashboard

The biggest measurement risk is that conversions are modeled. A conversion credited to Google may have happened anyway, the same event may be claimed by several networks, and modeled results cannot be audited or deduplicated externally. Cohorts from different placements may also pay and retain very differently, so platform-reported return can mislead.

A practical incrementality test is to isolate one country and platform, spend enough for the effect to be visible in total traffic, and compare overall subscriptions or revenue with and without the channel. If the channel is only about 1% of total subscriptions, the signal will be hard to detect. Use incrementality analysis or media-mix modeling to inform budget decisions, even though those methods are not precise enough for day-to-day campaign optimization.

Pre-registration campaigns follow the same logic with extra uncertainty: they bid on pre-registrations, stop at release, and some pre-registrations never become installs. Keep budgets modest, set countries carefully, and switch to standard campaigns at launch.

Frequently asked questions

What are Google app campaigns?

Google app campaigns, often shortened to GAC and formerly called Universal App Campaigns or UAC, are Google's automated app-promotion product for Android and iOS. Ads can appear across Google Search, YouTube, Play Store placements on Android, and the Google Display Network, largely through AdMob. You do not choose keywords, placements, audiences, or devices in the interface; Google's system decides delivery from your budget, bid, location, conversion signal, and creative assets.

What should I set up before launching?

Set up conversion tracking before launch, normally through Firebase or a mobile measurement partner, and link Play Console for Android. Choose an optimization event that is frequent enough to learn from but close enough to revenue or retention — for many subscription apps, trial starts are the practical middle ground. Then supply a varied asset set across text, images, video, and, where relevant, HTML5, because missing formats can close off placements such as YouTube.

How do Android and iOS campaigns differ?

Android has the Play Store search and display inventory and richer observable signals, which helps Google optimize. iOS lacks Play Store placements and, in the period covered here, Google did not use SKAdNetwork data to optimize campaigns, relying instead on modeled conversions extrapolated from engagement and limited observable events. That makes iOS results harder to audit and compare.

How should I judge whether results are real?

Treat reported conversions as directional, not as proof of incremental revenue. Modeled conversions cannot be cleanly deduplicated against other channels, and cohorts from different placements may retain and pay differently. Isolate a country or platform, spend enough to observe total traffic, and use incrementality analysis or media-mix modeling to check whether the channel moves the overall bottom line.

When do pre-registration campaigns make sense?

Pre-registration campaigns bid on pre-registrations instead of installs or in-app events, stop at release, and require careful country selection. A pre-registration may never become an install, so long pre-registration windows carry extra risk. They fit mainly large brands with outside demand or teams that need to test store assets before launch, rather than most new apps.