Performance Max gets described as automation you cannot steer. That framing is convenient, and it is mostly wrong. The campaign type hides reporting detail, but it does not hide the levers.
Asset groups are the real targeting
A single asset group holding every product and service is the most common cause of poor results. The system cannot separate intent it was never shown. Splitting by service line, margin, or buying stage gives the model a boundary to optimise inside, and it makes the reporting you do get legible.
Audience signals are a starting point, not a cage
Signals do not restrict delivery. They tell the system where to begin looking. Feeding it your actual customer list, your converters, and high-intent search themes shortens the learning period considerably. Feeding it a broad interest category tells it almost nothing.
Conversion quality decides everything downstream
If every form fill counts as one conversion, the campaign will optimise toward whatever produces the most form fills, which is usually the least qualified traffic. Values that reflect real pipeline outcomes change what the model chases. This is the single highest-leverage change available in most accounts, and it is a tracking task rather than a campaign task.
Exclusions still work
Brand exclusions, negative keyword lists at account level, and placement exclusions all apply. Using none of them and then concluding the campaign type is uncontrollable is a common sequence.
Performance Max rewards accounts that give it structure and honest conversion data. It punishes accounts that give it neither, which is why results vary so widely between advertisers running what looks like the same campaign.
