Performance Max Is Not a Replacement for Search.

Performance Max is a complementary campaign surface, not a Search replacement. Treating it as a Search successor destroys signal, cannibalises the branded baseline, and reduces the operator's ability to read paid-channel performance against revenue.

The claim has two parts. The first part is empirical: in every PMax-replaces-Search account the firm has read, the same three failure modes appear. Branded-Search impression share drops because PMax has no exclusion. Asset-group structure obscures where spend is going. Reported conversions rise while recognized purchase revenue stagnates. The second part is structural: PMax and Search read different parts of the auction. Collapsing the two into one campaign type does not simplify the account; it removes the legibility the operator depends on to make decisions.

The position is not "do not run PMax." The position is run PMax under four configured conditions, alongside Search, with reporting that reconciles the two surfaces against backend revenue.

The structural argument is that PMax inherits Search's converting traffic for cheap and reports it as PMax-attributed. Without brand exclusions, the algorithm finds branded queries inside the served-traffic mix and indexes there because branded queries are the cheapest source of conversions in any account that has a recognizable brand. The operator's "PMax wins" are the operator's branded baseline being repackaged at a higher CPC and credited to the new campaign.

Five failure modes follow.

Failure mode one. Branded-Search cannibalisation. Without account-level brand exclusions, PMax serves on branded queries that were already converting through Search at lower cost. Reported PMax conversions rise; reported Search conversions fall; total revenue does not move. The case file The PMax Campaign That Ate Its Own Branded Search documents the pattern in detail.

Failure mode two. Asset-group opacity. PMax campaigns have one or more asset groups, but the platform does not expose served-query-level data the way Search does. Operators lose the ability to inspect what queries spend is going against. The reporting layer narrows. Decisions that used to be made on search-terms data now get made on aggregate ROAS, which is a different type of signal.

Failure mode three. Conversion-goal corruption amplification. PMax is more sensitive to conversion-goal definition than Search is because it has fewer levers. A corrupted conversion goal — add-to-cart counted as primary, value parameter attached to the wrong event, attribution model misaligned — produces a worse outcome inside PMax than inside Search, and the case file The Conversion Goal That Counted Add-to-Carts as Sales documents how that compounds over twelve weeks.

Failure mode four. Reporting attribution loss. Channel-level reads collapse when one campaign type covers six platform surfaces. The operator can no longer say "Search produced X dollars" or "Display produced Y dollars" because PMax served on both. The ability to read channel performance against revenue, which is the input to every subsequent budget decision, is lost.

Failure mode five. Optimization toward existing customers. PMax with first-party customer-list audience signals trains the algorithm to find the people most likely to convert, which is almost always existing customers. The campaign acquires nothing new. Existing-customer revenue gets reattributed to PMax. New-customer acquisition stalls because the campaign was never instructed to find new customers, and the operator does not see the stall in the dashboard because PMax-reported revenue keeps rising.

Below is the working spec. Each condition has three numbered moves. The moves are read in order, completed in writing, and signed before the next condition is configured. The whole installation completes in under twenty hours of operator time on a typical Shopify account, regardless of campaign count.

Every methodology has assumptions. Naming the assumptions is part of defending the position. The four conditions assume correctly-configured Merchant Center feeds and trustworthy GA4 attribution. When either assumption is false, the read needs a different first move.

The Reference pages in the Performance Max cluster, the Reference pages in the Google Ads Waste cluster, the case files this position was written against, the companion position, and the hub. The graph below is the cluster map.