Performance Max has controls now, so a bad build is yours

Performance Max is not a black box any more, and the refusal to run it on transparency grounds has run out of road. Brand exclusions, campaign-level negative keywords up to 10,000 entries, account-level placement exclusions from January 2026, and channel-level reporting between them cover almost every objection that was legitimate two years ago.

Which changes who is responsible when a PMax campaign underperforms. It is now a build problem, and the build is yours.

What you can actually control

Worth listing plainly, because a lot of teams are working from a 2023 mental model.

Brand exclusions. You can stop the campaign serving on your own brand terms, which is the single most common cause of a PMax campaign that looks brilliant and is actually harvesting demand your brand campaign already owned.

Campaign-level negative keywords, up to 10,000 per campaign. This was the original headline objection and it is resolved. Job seekers, support queries, wrong product lines, all of it can be excluded at the campaign level without asking anyone.

Account-level placement exclusions, available from January 2026. This is the durable one. Placement lists maintained at campaign level get lost in migrations and rebuilds. At account level they persist, which makes brand safety a policy rather than a chore repeated per campaign.

Channel reporting. You can see performance split by channel rather than one blended number, which means you can finally answer whether the campaign is a Shopping campaign wearing a costume or genuinely finding new demand.

Asset group and search term reporting. Less granular than Search, sufficient to diagnose.

What you still cannot control

Be honest about the gap, because it is real.

You can see channel performance, but you cannot fully steer channel allocation. If the system decides your budget is best spent on Display, your levers are indirect: asset quality, exclusions, and the conversion signal you feed it.

Placement visibility is better than it was and still not complete. You get enough to exclude the obvious waste, not enough to build a whitelist.

And the campaign will still absorb credit for demand generated elsewhere unless you structure against it.

The build that works

Four decisions carry most of the outcome.

Separate asset groups by real commercial difference, not by taxonomy. Margin tier, buying cycle, or audience. Not Products A to M and Products N to Z. The asset group is the unit the system uses to understand what it is selling; if the grouping is arbitrary, the signals inside it contradict each other.

Make the feed the primary lever if you have one. Titles, product types and custom labels do more work than any setting in the interface. Custom labels for margin band, stock depth and seasonality let you build campaigns around contribution rather than revenue, which is the difference between a PMax campaign that grows the business and one that grows the top line while margin falls.

Exclude your own brand where a brand campaign exists. Then compare the two campaigns honestly. If PMax volume drops sharply on brand exclusion and total brand volume holds, you have just found out what it was really doing.

Optimise on conversion value, not conversions. PMax with a raw conversion target will find you the cheapest conversion, which in a multi-product business is almost never the one you want.

Where a Search and Shopping pair still wins

It would be dishonest to say PMax is the right answer everywhere.

A small, clean catalogue with strong query control and an experienced buyer still often performs better on a well-built Shopping campaign paired with a tightly managed Search campaign. The reason is simple: with a hundred SKUs and a clear query set, a human can allocate better than an automated system with limited feedback, because the human knows which SKUs carry margin and which carry warranty risk.

The crossover happens with catalogue size and query diversity. Somewhere past a few thousand SKUs, or a service business with dozens of genuine use cases, the manual approach stops being able to cover the surface area and the automation wins on reach even where it loses on precision.

Decide which side of that line you are on before choosing the campaign type, rather than after.

The diagnostic when it underperforms

Run these in order, because they cascade.

1. 1.Check brand exclusion status. Half of all PMax disappointment is an attribution illusion.
2. 2.Check channel split. If 70% of spend is on one channel, the campaign is that channel with extra steps, and you should compare it to a native campaign of that type.
3. 3.Check the conversion action set. Value-based bidding on a broken value signal produces confident nonsense.
4. 4.Check asset group coverage. Thin asset groups get starved; the system will concentrate on whichever group has the deepest creative.
5. 5.Only then look at the budget and target.

Most teams start at step five, which is why the same conversation happens every quarter.

The controls exist now, so the honest question is no longer whether Google will let you see inside the campaign, it is whether anyone in your team has been given the time to build it properly.