Should you trust Google Performance Max campaigns?

What does Performance Max control, and what do you?

Google decides how one budget moves across seven of its own surfaces and what it bids in each auction. What stays on your side is everything upstream of that.

Those seven surfaces are the ones Google named when it added channel level Performance Max reporting in late January 2026, according to Refonte Learning's August 13, 2026 write-up. Before that, the spend went into one line. On the accounts we run, the inputs we still hold are the budget, the conversion action we optimize for, the product feed, and the creative we upload. Everything after that is Google's call.

That division matters for how you read the results. Lyra's State of Google Ads report flags that Performance Max adoption skews toward accounts that already have strong product feeds and working ecommerce fundamentals. The feed is the part you own. When it is thin, Performance Max has less to work with, and the campaign type takes the blame.

Do its returns hold up against other campaign types?

In the incrementality data we have seen, yes, in the middle of the pack. Cassandra's marketing mix model study puts Performance Max at 4.64x median incremental ROI, under Search Non-Brand and over Search Brand.

The study, published September 3, 2026, covers 59 advertisers, 253 models and 383 million dollars in tracked spend from 2023 to 2025. Its finding on spread is the more useful half. Performance Max results clustered tighter than either Search type, so the median sits closer to what a given account actually gets.

Then there is the number you will see quoted everywhere. Lyra's report on 94 Google Ads accounts (522 campaigns, 3,014,680 dollars in tracked spend between June 2025 and April 2026) puts Performance Max at 9.32x reported ROAS against 2.61x for Search. That is platform attribution, and Lyra itself flags the selection bias in who adopts Performance Max. Some of those conversions were headed to Search anyway. Reported ROAS counts them once for Performance Max, and an incrementality model strips them out.

What do advertisers complain about most?

Where the money lands. Brand and agency executives told Digiday that Performance Max spend often ends up on lower quality inventory.

One of them checked Google Analytics 4 directly and estimated that more than half of their traffic came from outside the geographies they had targeted. Digiday published that on May 11, 2026.

We do not know how common it is. One executive, one account, and no study we can point to that measures geographic leakage across a sample the size of Cassandra's 59 advertisers.

It is testable on your own account, though. Pull the geography report in GA4, compare it against the countries set in the campaign, and you will at least know whether the complaint applies to you.

Did Google's new reporting fix the visibility problem?

Partly. Google Ads API v23 and channel level reporting landed in late January 2026, so you can see which surfaces spent your budget, and you still cannot tell Google how to divide it.

Refonte Learning dates the rollout to late January 2026 and notes it followed more than a year of advertisers asking where their Performance Max money actually went. The seven surfaces now reported separately are Search, Search Partners, Gmail, YouTube, Display, Discover and Maps. That is a real change. It is also a read-only one.

Seeing that most of a budget went to Display does not let you move that budget to Search. The update added no per channel allocation control, so the levers are the same two as before, which are whether to run the campaign type at all and how much to give it. Our read is that the reporting is worth having for the argument it settles internally, and that it changes nothing about what you can do on Monday morning.

The reports name the surface that spent the money, not whether the visit came from a country you picked, and nothing we read breaks 4.64x down by channel. That leaves us sizing Performance Max as a steady middle performer and checking geography by hand. A channel level incrementality number would settle it. We have not seen one.