Performance Max black box: what it decides, what you still control, how to read results

Google's help page for PMax lists brand exclusions, negative keywords, search themes and final URL expansion as controls you can use¹. In plain English:

• Brand exclusions. Keep the campaign off searches for your own brand names. This matters for the next section.
• Negative keywords. Block searches you do not want to pay for.
• Search themes. Give the campaign hints about the searches you care about. They guide. They do not lock.
• Final URL expansion. Decide whether Google may send traffic to pages beyond the ones you named, or only to those you chose.

Beyond those, the main controls sit outside the campaign:

• The conversion goal and its value. The campaign works towards what you give it. A wrong goal scales the wrong thing. Choose your bid strategy with the margin in mind, not only revenue. A target ROAS measures revenue, not profit.
• The product feed. Titles, prices and product groups steer what shows. A weak feed gives weak results, however clever the bidding.
• Budget and the product mix. If the best seller eats the budget, the rest of the catalogue gets little. That is a feed and structure question.

Google's help page lists asset, asset group, channel performance and placement reporting for PMax¹. That is more than the black-box reputation suggests. Visibility is still a recurring theme in the trade: there are whole retail Q&As on PMax² and agency write-ups of its results and concerns³. Those are opinions, not a survey.

Reporting inside Google Ads changes over time, so check what your own account shows today. Write down what you can see and what you cannot. Do that before you judge the results. If you cannot see where the money went, say so in the report. Do not fill the gap with a guess.

Whatever the account shows, it shows Google's attribution of the sales. Reported ROAS is not incremental ROAS (iROAS), which is revenue caused by the ads divided by spend.

Brand searches tend to convert well, because the searcher already knows you, and they flatter a campaign's average. If PMax absorbs them, its reported result goes up while total sales may not. That is cannibalisation: paid activity taking credit for sales that would have arrived anyway.

How to check:

1. Look at what your account shows. Review whatever search term or category views are available for the campaign. If brand terms appear, note how much of the spend they take.
2. Compare brand search before and after PMax started. Did your separate branded search campaign lose impressions? Check impression share for brand terms. A fall after PMax went live is a clue, not proof.
3. Apply brand exclusions deliberately. Then watch total brand-driven sales, organic and paid together, not only the PMax line.
4. Use the shop as the referee. If PMax reported revenue rises and shop revenue is flat, you have a measurement question, not a growth story.

Branded search has its own test logic. Read is branded search incremental before you decide what to do with it.

Some practitioners run PMax alongside Standard Shopping ads campaigns. A trade-press article describes this hybrid approach for ecommerce⁴. We do not treat that as proof it works for your account.

Our view on when a hybrid is worth trying:

• You need more control over which products get budget, and PMax keeps drifting to best sellers.
• You want to protect a set of products or a margin tier.
• You want a comparison group for the test below.

The cost is a more complicated account. Two campaigns can compete for the same searches. Plan how you will read the results before you build it.

Google has a help page on PMax experiments⁵. Use it when it fits your setup. The logic is the same as for any incrementality question: what would have happened without the campaign?

Three ways, from cheapest to hardest:

1. A before-and-after read at the shop level, with other changes written down. Weak, but better than the dashboard.
2. A platform experiment, splitting traffic between setups.
3. A holdout, for example regions where the campaign is off, if volume allows.

Each has a bias. Pre/post is hit by seasonality and promotions. Regions differ. State the bias in the report. If your volume is low, a cheaper method may be the honest choice.

• Which goal and value does the campaign optimise towards, and why?
• Are brand terms excluded? If not, how much of the spend do they take?
• What can we not see in the reporting? What are we assuming instead?
• How would we know whether PMax added sales, rather than collected them?
• What changes if we turn it off for a region?

For ZFD: Our view is that PMax is a tool with documented controls and an unproven extra: set the goal from your margin, exclude what should not be bought, and check the result against the shop. For more on the platform, see Google Ads for ecommerce.