Why is my Google App campaign spending on YouTube? Signal and control are two different things.

An iOS App campaign can move a large share of its spend from Search into YouTube, on days you did not choose, while Search still looks like the better network in your reports. Google documents the mechanism. What it does not document is any link between the measurement stack it asks you to build and where the campaign then spends. Those are two different things, and mixing them up leads to the wrong fix.

If you only read one paragraph: App campaigns serve across Search, Google Play, YouTube, Discover, Display and AdMob, and Google’s system decides the mix. You cannot make one Search only, and you cannot exclude YouTube as a network. Better iOS measurement (Firebase, ICM, on device measurement) gives Google more conversions to learn from, which is good, but it does not give you more say over the mix. If Search is the part of the campaign you value, buy Search with a product built for it, a standard Search campaign or a web to app funnel, and judge the App campaign on its blended, matured numbers.

Google renamed Universal App Campaigns to App campaigns in 2019. Most people I work with still say UAC.

What does the Search to YouTube shift look like in practice?

A UA lead in a private community I am part of described it well. Through 2025, their iOS App campaign spent most of its budget on Search and performed well. To test tROAS, they built the full measurement stack Google asks for: Firebase, the ATT prompt, Integrated Conversion Measurement and on device measurement. After that, YouTube began taking a large share of spend on some days, results fell on those days, and the weeks where Search stayed on top kept performing. tROAS never reached scale on their revenue signal, so they went back to tCPA on a deeper event. Their Google rep suggested better video creative and Search only web to app campaigns.

Their Android campaigns were fine throughout, on longstanding tROAS. I am not naming the person or the app, and I have not seen the account. Read it as a well described pattern, and the rest of this piece as what Google’s documentation and my own test say about it.

Why does Google move App campaign spend from Search to YouTube?

Because App campaigns are built to spend across networks, and Google says so. Its page on App campaign network distribution lists Search, YouTube, Google Play, Discover, the Display Network and AdMob, and says the system distributes delivery to drive conversions based on your goals, budget, targeting and assets, using a prediction of how likely each user is to convert.

The same page explains the part that frustrates people. A network can look expensive next to another, but the cheaper one might not have the inventory to serve your campaign, so the system pursues traffic on the networks where you are more likely to get conversions. That is close to what happened in the case above: Search was the best network, Search ran out, and the campaign kept spending.

Search having the best CPA in your report does not mean Google could have spent the whole budget there. The missing number in most network discussions is how much more Search inventory existed at an acceptable marginal cost, and the interface does not show you that.

Did better measurement cause it?

Nothing in Google’s documentation says so. Google describes Integrated Conversion Measurement as event level, close to real time reporting inside App Attribution Partners, and says on device conversion measurement increases the number of iOS conversions it can observe and improves optimization. Neither page mentions network allocation.

So what the story shows is a sequence. It does not show a cause. The account gave Google more usable conversions. Google’s bidding got more confident about who converts. The campaign kept looking for conversions wherever it could find them, and Search was already saturated. The measurement upgrade may have made the system better at the job you asked it to do, while the job you actually wanted was “buy Search.”

That is the core of it. Measurement decides how much Google knows. Campaign type decides how much you control. Improving the first does not move the second.

Why can Search win and still fail to scale?

Because intent is finite. Someone typing a query has told Google what they want. Someone watching a video has not. Search converts better for most apps I have seen, and there are only so many relevant searches a day, and other advertisers are bidding on them, including your own standard Search campaigns if you run any.

My own test is a useful check on the “Search is always best” story, because it did not hold there. Across $73K on six campaign types, Search took 58% of the budget and returned 25%, Display took 32% and returned 43%, and YouTube took 9% and returned 13%. YouTube was the weakest network per dollar in my account too, but Search was not the one I would have locked the budget into. Display was. The full breakdown is here, including why Max Conversion Value beat tROAS in that test.

The network mix decides the campaign’s result, you do not set the mix, and so you need to see it at all times. Google’s own iOS troubleshooting page tells you to segment the campaign by network when performance shifts.

Can you make a Google App campaign Search only?

No. Google documents no Search only setting. You can exclude specific placements, meaning individual sites, apps, videos and channels, and the same page says you cannot exclude all of youtube.com as a placement exclusion at the campaign level. A text only App campaign leans toward Search, and Burak and I explain in our Search piece why it still does not guarantee Search only delivery.

The placement report will not rescue you either. Google’s page on App campaign placement reports says impressions are the only performance metric available in them, and low volume placements roll up into an “other” line. You can see where the impressions went. You cannot see what each placement earned.

If Search only is the goal, you are asking an App campaign to be a different product. Google describes Search campaigns as keyword driven, where you choose the terms. That is the product that gives you the control.

Does poor YouTube performance mean your video is bad?

Not on its own. The rep’s answer in the case above was better video, and I understand why they say it, but it is one of several explanations and the least measurable one. A YouTube heavy week can underperform because the intent is lower, because the bidding model chose a different audience, because your optimization event is thin, because of conversion lag, or because the video was never made for that placement.

The last one is worth checking before you argue about the rest. Google says that if you do not provide video, it may make a video ad for you from your other assets and your store listing. If your Search ads run on copy you wrote and your YouTube ads run on a video Google assembled, the comparison between the two networks was never fair. Most subscription teams I know test creative on Meta and port the winners to Google, which is sensible, and it means the Google video is usually a Meta ad in a Google placement. That is a creative question, and it sits after the mix question, not before it.

Why is this harder to read on iOS?

Because iOS conversion data arrives in pieces, on different clocks. Google’s page on iOS App campaign measurement lists three views: Google Ads modeled conversion reporting, ICM reporting inside your attribution partner, and SKAdNetwork. The modeled reporting is subject to delays of up to five days, and the page says the ICM data is not available in Google Ads reporting today.

On Apple’s side, AdAttributionKit, the newer framework, runs alongside SKAdNetwork, and how much detail a postback carries depends on privacy tiers. None of those datasets is the one you see in Ads Manager on Meta.

So when someone says YouTube was terrible yesterday, yesterday is not finished. The closer your optimization event is to revenue, the more that delay matters, which is why Google tells you to evaluate tROAS over a 14 to 30 day window and to leave out the most recent days.

Should an iOS subscription app use tCPA or tROAS?

Neither by default. The question is which one gives Google a signal that is both close to money and frequent enough to learn from.

Google’s rules make the choice for many accounts. tROAS on App campaigns requires the Google Analytics for Firebase SDK. The tROAS eligibility page asks for at least 10 conversions a day, or 300 in 30 days, and recommends running a tCPA campaign before a tROAS one. Google’s setup guidance puts the switch at a stable volume, typically 30 or more conversions in 30 days, with three to four weeks of value history behind it. Since August 2025, Google also offers maximize conversions bidding on App campaigns and tROAS on iOS.

For a subscription app with sparse purchases, a tCPA campaign on a deep event, a trial that survives a cancellation window or a first payment, is often easier for Google to learn from than a value signal it barely receives. The case above ended up there for that reason. The event has to mean something, though. A high volume event that does not predict revenue gives the algorithm a clean target and the wrong users. I wrote about which event to pick and when value bidding pays off, and both apply here even though the second piece is about Meta.

In my own test, Max Conversion Value beat tROAS on both platforms in every market. My sequence is still to start without a target, wait 10 to 14 days, and add tROAS only when I need to hold a return and have the budget to feed it.

Which product gives you control over Search?

Three routes buy Search demand for an app, and they trade control for friction in different amounts.

A standard Search campaign gives you the keywords and the network, and on iOS it loses the click ID at the App Store, which is the gap Burak and I mapped, with the workarounds we know of. Google has started to close it: its Web to App Acquisition Measurement, announced in September 2025, attributes iOS installs from Search and Shopping web campaigns to one campaign instead of reporting them twice, and Google’s 2026 announcements extend ICM to web to app journeys and to EEA, UK and Swiss users.

A web to app funnel moves the first conversion onto a page you own, a landing page or a quiz that registers or subscribes the user before the store. You get keywords, landing pages, first party events and the funnel steps. You also add a page between the ad and the app, and for a subscription app that page can carry the payment, which is why teams like the one in the case are expanding it. Treat it as its own acquisition system with its own economics.

Running App campaigns and Search or web to app side by side is fine. Judging them as one thing is not. Auction overlap between them is possible and does not prove cannibalization; only a deliberate change in exposure, with total acquisition measured, tells you that.

What should you check when YouTube takes over?

In this order, because each step can end the investigation.

1. Did the mix change? Segment by network and put a number on the Search, YouTube and Display shares before and after.
2. Has the data matured? Do not read the last five days on iOS as if they were final.
3. Did the target change? Event, tCPA or tROAS value, conversion window, anything you touched in the same weeks as the measurement work.
4. Is the creative fair across networks? Compare the video you supplied with anything Google generated.
5. Did Search hit its ceiling? Launch a separate Search or web to app campaign and see whether profitable Search demand was left on the table.
6. Does it repeat? One bad week is an observation. The same relationship under a controlled change is evidence.

What would I test next?

I would stop trying to make the App campaign behave like a Search campaign. Give it clean measurement and an event that means money, and read it on matured cohorts at the account level, the way I read any Google account. Then buy Search separately, with a standard Search campaign where iOS measurement allows it, or with a web to app funnel where it does not, and compare the two systems on business outcomes.

The question stops being how to get UAC back onto Search. It becomes which parts of your acquisition benefit from Google’s automation, and which parts are valuable enough that you want your hands on them. That is a better experiment, and it is the one better measurement was never going to run for you.

The Google and Apple pages linked above were checked on September 26, 2026, and Google says its iOS measurement will keep changing. The $100K figures are one account’s numbers, not a controlled experiment. The case is one team’s description of their own account, not something I audited.