PMax reports more purchases while branded Search reports fewer. Total shop orders barely move. The campaign may be taking credit for existing demand, adding valuable reach, or doing both. Its own ROAS cannot distinguish those explanations.
Our recommendation is to define the business comparison before excluding the brand. The question is whether the chosen campaign mix produces additional useful outcomes at an acceptable cost. Moving a purchase between campaign reports is a different result from gaining an additional customer.
Write the question before touching the setting
Two common questions require different comparisons. “Does adding PMax help our existing mix?” asks about incremental campaign value. “What happens if this PMax campaign cannot serve selected brand queries?” asks about a restriction within the mix. An exclusion-only comparison does not automatically answer whether all PMax activity is incremental.
Write down the treatment, control, population and decision. For a shop, choose retained orders or contribution after relevant variable costs. For B2B, use a consistently defined accepted enquiry or later CRM outcome. Do not change the conversion definition halfway through the investigation.
Also distinguish your retailer’s name from the product brands you sell. Excluding a manufacturer can remove commercially useful product demand beyond searches for your own business. A vague instruction to “exclude brand” is not enough for implementation.
Map demand and measurement before the comparison
Inspect available search-term information, campaign reports, landing routes and the order system. Record how you classify your name, spelling variations and mixed brand/category queries. A limited visible query report does not identify every interaction or classify every buyer.
Define a new customer using an agreed purchase-history rule and source. A person searching the brand may be buying for the first time; a generic query may belong to an existing customer. New-customer classification and non-brand traffic are useful dimensions, but neither measures what would have happened without the campaign.
Align value, currency, attribution settings and conversion delays. Deduplicate business orders using stable order identifiers. Do not add Google Ads and GA4 attributed conversions as if they were separate orders. For leads, reconcile qualification and sale dates as well as submission dates.
Before launch, record promotions, stock limits, other media and planned site changes. These are potential explanations for a changed result. If measurement is unreliable, repair it before buying a comparison that cannot be interpreted.
Understand what Brand Exclusions remove
Google’s current instructions limit PMax brand exclusions to Search, Shopping and YouTube search inventory. An option can retain Shopping ads for excluded brands. This is not a brand exclusion across every placement or every interaction with a known customer.
Record the exact brand list and Shopping choice before the test. Verify which business and manufacturer names are covered. Keep separate branded Search campaigns and their spending visible, since demand may move into them after a restriction.
An audience signal is not a sealed audience, and a new-customer goal is not an incrementality experiment. Likewise, a lower PMax ROAS after an exclusion can simply reflect a different mix of attributed conversions. It does not establish an improved or damaged business on its own. Our general PMax guide covers campaign setup and controls.
Choose a comparison that can answer your question
Google distinguishes PMax experiment types. The uplift option without Merchant Center and a Shopping upgrade experiment address different configurations. Check eligibility and the current setup available in your account; do not assume that every shop can use the same test.
For an eligible uplift experiment, inspect treatment, control and the comparable campaigns. Google’s uplift instructions state that comparable campaigns are excluded from reporting by default. Enable the view with comparable campaigns and inspect the membership before drawing a conclusion about the combined result.
When a suitable platform experiment is unavailable, a carefully designed geographic holdout may be an option. It needs enough comparable regions, control of contamination and adequate observations. Sequential before-and-after periods are weaker: season, brand demand and competitors may change between them. Label that result as directional.
Agree exposure and stopping rules before launch. Keep the relevant offers, measurement and competing interventions stable. Select a duration and sample plan that account for conversion lag and the smallest business difference worth detecting. Avoid stopping simply when the dashboard first shows a favourable result.
Read the combined result, not the winning campaign
Consider a hypothetical equal-scope comparison, not a client case or a statistically established test. Under condition A, PMax receives credit for 120 orders and other Google campaigns for 80. Under B, PMax receives 70 and the others 130. Each condition therefore reports 200 in total.
The 50-order shift does not establish an extra business order. Check deduplicated shop orders, other channels and costs before interpreting even that total. If A costs €4,000 and B €3,500 for the same retained contribution before advertising, B leaves €500 more under these assumptions. That is arithmetic, not proof that excluding brand will save the same amount.
| Result across the agreed comparison | Interpretation to investigate | Decision supported |
|---|---|---|
| PMax falls, other campaigns rise, business outcome stays stable | Attribution or demand has shifted | Compare total cost and contribution |
| Combined retained contribution rises beyond extra spend | Additional value may exist in this population | Check uncertainty and operating limits before expanding |
| Platform purchases rise, retained orders do not | Goals, duplicates, timing or cancellations may differ | Reconcile measurement before scaling |
| Direction changes with small samples or overlapping uncertainty | Evidence does not resolve the question | Keep the result inconclusive or redesign the comparison |
Compare both the point estimate and uncertainty. A positive estimate with a wide interval can still be compatible with an unacceptable loss. Specify how much confidence the spending decision requires, rather than treating every positive number as permission to scale.
Decide what the business will do afterwards
Accept, reject or retain the current structure according to the pre-agreed rule. A test can also finish without a decision. Document the limits instead of repeatedly extending it until a preferred answer appears.
Keep the brand restriction if it serves an explicit commercial objective and the overall evidence supports it. Remove or alter it only against the same business criteria. There is no universal requirement that every PMax campaign exclude brand or retain it.
Canem Errant’s Google Ads service can review goals, brand activity and the route from an attributed conversion to a useful outcome. Begin with one comparison, its eligible population and maximum spending exposure. The deliverable is a reasoned campaign decision, including an inconclusive outcome when the data requires it.
Frequently Asked Questions
Does a high share of brand queries prove cannibalization?
It shows brand activity worth investigating. Cannibalization concerns outcomes that would have happened through another route; identifying that difference needs a suitable comparison.
Does excluding brand make every PMax sale incremental?
No. The exclusion has inventory limits and does not identify the counterfactual outcome. Non-brand demand can also include existing customers or people who would have purchased anyway.
What if the account is too small for a useful experiment?
Use a bounded diagnostic, reconcile orders and costs, and keep conclusions directional. Do not present an underpowered comparison as proof that PMax works or fails.
