Performance Max is the campaign type Google has been pushing every account toward since it replaced Smart Shopping. The argument about whether it “works” is badly framed: it works very well for some accounts and conceals serious problems in others, and which one you have is fairly predictable in advance.
What it is, in one line
A single campaign that serves across Search, Shopping, YouTube, Display, Discover, Gmail and Maps, where you don’t choose the placement. Instead of ad groups with keywords, you supply asset groups — headlines, descriptions, images, video, and if you sell products, the Merchant Center feed. The algorithm decides the rest.
That last sentence is the entire conversation.
The problem almost nobody checks: brand cannibalisation
This is the most important part of this post.
Performance Max can serve against searches for your own brand. Brand searches convert better than anything else in any account, because the person already knows you and already decided. When PMax absorbs them, three things happen at once:
- The campaign’s ROAS goes up, because it is claiming conversions you were going to get anyway.
- Your branded Search campaign’s traffic falls, making it look like that campaign stopped working.
- Account-level ROAS doesn’t move, or gets slightly worse.
You see a campaign with a spectacular return and conclude it deserves more budget. What’s actually happening is that an expensive channel is claiming credit for demand that already existed. It’s the same attribution problem as always, except now it’s happening inside a single campaign where you can’t see the split.
How to check. Look at traffic and conversions for your branded Search campaign before and after PMax went live, and look at the account total rather than the campaign total. If PMax went up and brand went down in roughly the same proportion, you gained nothing and paid more for it.
How to fix it. Exclude your brand terms from PMax using campaign- or account-level brand exclusions, depending on what your account has enabled, and keep brand in its own Search campaign where you control the bid and the message.
Where Performance Max genuinely earns its place
- Ecommerce with a large catalogue and a well-built feed. This is what it was designed for. With good titles, good images and complete attributes, PMax reaches long-tail demand that would be impossible to build manually.
- Accounts with high conversion volume. The model needs data. Below roughly 30 conversions a month it’s guessing, and you’re paying for its education.
- Mature accounts where Search is already well covered. PMax as a layer on top of a disciplined Search structure works. PMax as a replacement for structure is how accounts lose visibility into their own performance.
- Markets or segments you can’t staff. Regional accounts with fifteen markets and one person: PMax genuinely buys coverage that manual campaign management can’t.
Where it hides the problem
- Low conversion volume. Not enough signal, so it optimises toward whatever converts easily — usually your existing customers and your brand.
- Lead gen with unqualified leads. PMax will efficiently generate the cheapest version of whatever you told it to optimise for. If your conversion is a form fill and you don’t feed back lead quality, it will find you a lot of very cheap bad leads. See lead quality in Google Ads.
- Broken or shallow conversion tracking. Every automated campaign type amplifies whatever your measurement says. If the measurement is wrong, PMax is wrong faster and with more of your money.
- Accounts where nobody looks at the asset group report. PMax’s reporting has improved a lot, but it still requires you to go looking. Accounts that set it and forget it drift.
What changed recently, and what it means
Two updates worth knowing about as of late 2026:
Destination choice for local advertisers. PMax now surfaces website and Business Profile as destination options. Useful if you’re driving store visits or calls — but verify how calls, forms and visits are being measured before you switch, because the conversion definitions differ and your historical comparison will break.
Product value optimisation is in selective beta, letting you prioritise product values for bidding rather than treating all revenue as equal. This is genuinely valuable for catalogues with uneven margins — it’s the closest thing to bidding on profit instead of revenue. Check eligibility before planning around it; access is not universal.
Neither of these changes the underlying trade: you gain reach and lose control, and the accounts that do well are the ones that keep enough structure outside PMax to notice when something goes wrong.
A structure that works
For most ecommerce accounts I’d run:
- Branded Search, exact match, manual control. Cheap, high-converting, and yours. Never inside PMax.
- Non-branded Search on your known high-intent terms. Where you can see the search terms and act on them.
- PMax for catalogue coverage, with brand excluded, a properly maintained feed, and asset groups split by product category or margin band rather than one group for everything.
- A measurement layer that isn’t the platform. Blended efficiency month over month, so you can tell whether the account is actually growing. See how to calculate ROAS.
Point four is the one that makes the rest safe. PMax’s own numbers will almost always look good. The question is whether the business number moved, and only a number the platform doesn’t control can answer that.
The short version
Performance Max is a good tool for catalogue coverage in accounts that already have volume, clean tracking and a disciplined Search structure underneath it. It’s a bad tool for accounts with thin conversion data, unqualified lead flows or broken measurement — not because it performs badly, but because it performs plausibly while telling you a story about branded demand you already owned.
Exclude brand. Watch the account total, not the campaign. Feed it good data or don’t feed it at all.
If your PMax campaign shows the best ROAS in the account and total revenue hasn’t moved, that’s usually the first thing I check.