Performance Max: what it is good at, and what it hides
The problem with Performance Max is not that it performs badly. It is that when it performs badly you cannot see why, and when it performs well you cannot tell what did it.
Performance Max is one campaign that buys across everything Google owns — search, shopping, display, YouTube, Gmail, maps — with the split between them decided by the model rather than by you. You supply assets, a feed if you have one, a conversion goal and a budget. It does the rest.
That description is also the whole argument, in both directions.
What it is genuinely good at
Finding demand you would not have queried. A keyword list can only contain what somebody thought of. PMax reaches people through signals no search campaign structure would have expressed, and in categories with broad, badly-articulated demand this is a real and repeatable gain.
Scale without proportional management. Once it has enough conversion history, it holds its ratio at budgets that would take several hand-built campaigns and a person watching them. For a small team this is the honest appeal, and it is not nothing.
Shopping inventory. Where there is a product feed, PMax is currently the most effective way to buy it. The feed does more work than the creative — a well-structured feed with real titles and attributes is the single biggest performance lever in the campaign, and most accounts treat it as a data export.
What it hides
It takes credit for demand you already had. This is the big one. PMax will happily serve people searching your brand name, convert them at an excellent rate, and report that excellence as its own. Those people were going to find you. If brand terms are not excluded and running separately, a good part of what looks like incremental performance is a re-labelling of traffic you were getting for free.
The reporting is thin by design. You get far less placement and search-term visibility than in a normal campaign. When cost per acquisition drifts, the usual diagnostic path — look at where it spent, find the leak, exclude it — is partly closed. You are left inferring from asset group performance and whatever the search term insights choose to show you.
It will buy things you would never have bought. Left alone, budget flows to the cheapest inventory that satisfies the goal, which frequently means display and video placements with impressive volume and unimpressive intent. In a restricted category that is not only wasteful, it is a policy exposure: you have less control over the context your ad appears in.
It optimises to whatever you told it. If the conversion goal is a form submission, it will find people who submit forms. Whether those people were qualified is a question the model neither asks nor can answer, and lead-generation accounts running PMax on raw form fills reliably produce large volumes of unusable leads at a beautiful reported cost.
When it works and when it does not
It works when the account already has a body of real conversion history, the goal reflects money rather than an intermediate action, the feed is clean, and brand traffic is fenced off. It works particularly well for e-commerce with a decent catalogue.
It does not work on a new account with thin data — a model trained on noise produces confident noise. It struggles where lead quality varies enormously, unless the CRM is sending outcomes back. And it deserves caution in restricted categories, where placement control matters more than reach.
How to run it so the reporting still says something
- Separate brand. Brand search in its own campaign, brand terms excluded from PMax. Without this you cannot measure anything else honestly.
- Set exclusions before launch, not after the first bad month. Placements, content types, and anything the category cannot appear next to.
- Split asset groups by something meaningful — product margin, service line, market — so that when the aggregate moves you can see which part moved.
- Feed the real goal. Offline conversion import where the sale completes elsewhere. If a lead is only valuable once qualified, send the qualification.
- Build the history on search first if the account is new. PMax is a scaling tool, not a starting one.
The honest summary
Performance Max is a capable buying mechanism attached to deliberately poor instrumentation. The capability is real and the opacity is not an accident — it is what allows the automation to work without arguing with you about every placement.
Run it where you can afford to trust it and have fenced off the ways it can flatter itself. Do not run it as a substitute for knowing what your account is doing.