Performance Max, two years in: what we’ve learned.

Two years on from the formal sunset of Smart Shopping, Performance Max is no longer a novelty product — for most of our retail and lead-gen clients it now carries the largest single share of paid spend. We have managed close to a hundred PMax campaigns since the migration window in mid-2023; this is the working summary of what we’ve learned.

One asset group per theme, not per product

The default mistake is to mirror your old Shopping campaign structure inside PMax: one asset group per product category. This rarely works in practice because PMax does not respect ad-group-style isolation the way classic Shopping does. Themes — price tier, audience signal, customer intent — are the right organising principle. We tend to start clients with three to five asset groups: high-intent searchers, broad prospecting, brand audience, and (where relevant) retargeting. Product feed segmentation through listing groups handles the SKU split inside the asset group.

Audience signals are not targeting — but they still matter

Asked over and over by clients: “If audience signals are only suggestions, why bother filling them in?” Because in our experience PMax converges materially faster when given strong starting signals. The empirical pattern: a campaign with high-quality customer-list audiences and detailed keyword themes hits target CPA in week two. A campaign with no signals takes four to six weeks and burns more spend on the way. Treat audience signals as the warm-start for the model, not as the targeting itself.

Search term insights, finally usable

The Search Terms Insights report has gone from useless (mid-2023, all aggregated into “Categories”) to genuinely actionable (late 2025, with downloadable category-level term lists). We now run a weekly sweep on Search Terms Insights and apply account-level negative keywords against any category whose conversion rate underperforms the campaign average by more than fifty percent. This single discipline has been the single biggest driver of cost-per-acquisition improvement we’ve seen in PMax.

Creative discipline matters more than copy variation

The temptation with Performance Max is to flood the asset library with every creative ever produced. This is the wrong instinct. PMax is an aggressive optimiser; given mediocre creative it will find the cheapest way to spend your budget on the cheapest impressions. Better: a small number of well-defined creative concepts, each with the full required asset count (15 headlines, 5 long headlines, 5 descriptions, 20 images at the right aspect ratios, multiple video orientations). Asset groups should look like distinct campaigns of their own, not a mixed library.

Brand exclusion is now a button, but it’s still essential

The brand exclusion feature shipped in Q2 2024 and is on by default for new accounts as of late 2025. If your account predates this you are probably still wasting spend on cheap brand traffic that PMax claims as conversion uplift. Always exclude your own brand from PMax. Always run a separate Brand Search campaign for the brand traffic. The conversion attribution becomes honest, and the unbranded performance metric becomes a real metric.

When to keep classic Search alongside PMax

We default to Search-plus-PMax for any client with meaningful brand search volume, regulated-industry compliance requirements, or a need for granular control over copy variants. PMax-only accounts work for high-velocity catalogue retailers with broad demand and minimal regulatory friction; they do not work well for B2B lead generation, high-consideration purchases, or any context where headline-level approval matters.

Reporting and the “black box” complaint

Yes, PMax reports less than classic Search. No, this has not stopped us reaching defensible conclusions about what’s working. Asset group reporting now exposes Conversions, Conversion value and ROAS at the group level. Asset-level performance ratings are workable, if blunt. Search Terms Insights gives directional intent data. Geo, time-of-day and device segmentation are available. There is no campaign in our portfolio for which a senior buyer can’t answer “what’s working and why?” using PMax’s native reporting alone.

What we still don’t do in PMax

Brand campaigns. Geo-restricted campaigns where the geo is narrower than a country. Campaigns where regulatory copy must be exact. Campaigns where measurement requires server-side conversion-action overrides that PMax doesn’t expose. We will probably continue running classic Search alongside PMax for these reasons for the foreseeable future.

Closing observation

PMax has gone from a beta experiment to a load-bearing campaign type in two years. The fundamentals of paid search — clear conversion definitions, honest creative, disciplined negative-keyword work, brand vs. non-brand separation — have not changed. The only thing that has changed is that doing them well now requires understanding how a black-box optimiser uses the inputs you give it. Treat the audience signals, asset structure and brand exclusion as if they were targeting decisions; that’s how you stay in control of a campaign type that gives you fewer levers than the one it replaced.