Performance Max has been Google Ads' default for ecommerce and lead-gen advertisers since 2023, and most accounts still aren't running it well in 2026. The category is too new for established best practice to be widely understood, the platform-native UI is opaque about what's actually being spent on, and the most-published 'PMax tips' content online is mostly outdated.
This guide is the operator's playbook for getting PMax to actually perform — what to optimize, what to ignore, and how to read the signals the platform does expose.
Asset-group strength is the primary lever
Performance Max measures every asset group on an internal 'strength' rating (Excellent, Good, Average, Poor) based on how complete the asset feed is. Asset groups rated Excellent get more impression share, lower CPMs, and better placement quality. The strength signals to chase:
- Minimum 15 images covering all aspect ratios (1:1, 1.91:1, 4:5 at minimum; ideally 4-5 of each)
- Minimum 5 video assets, including 6-second, 15-second, and 30-second variants
- Maximum-allowed headlines (15) and descriptions (5) using a mix of feature-led and benefit-led copy
- Search themes that match your actual top organic queries (not the brand-cannibalization keywords that distort the model)
Going from a 'Good' strength rating to 'Excellent' typically lifts the asset group's impression share 15-30% without changing budget or bid. It's the highest-ROI single optimization most PMax accounts have available.
Search themes need exclusions, not just adds
Search themes — Google's PMax targeting signal that tells the model what queries to bid on — are easy to add badly. The default mistake is adding broad themes like 'running shoes' or 'b2b software', which the model interprets as license to bid on any query containing those words, including low-intent comparison queries and brand-name queries that would have converted anyway.
The 2026 working pattern is using search themes sparingly (3-5 per asset group, narrow and intent-specific) and pairing them with brand-keyword exclusions at the campaign level. The brand-cannibalization exclusion is critical — without it, PMax cannibalizes existing brand-search campaigns, claims their conversions, and inflates reported ROAS while spending where you didn't need to spend.
What to monitor in the Insights tab
Google added a Performance Max Insights tab in 2023 that exposes some of what was previously hidden. The signals worth watching weekly:
- Search categories — what queries triggered ads. Look for cannibalization on brand terms and irrelevant query categories.
- Audience signals — which custom audiences and demographics the model is leaning into. Reweight if it's drifting from your target.
- Asset performance — which images, videos, and headlines are over-performing. Promote the winners; replace the laggards.
- Customer journey — view-through and click-through path attribution within Google's ecosystem.
When to split a PMax campaign
The strong default is one PMax campaign per business goal, not one PMax campaign per product line. Splitting too narrowly starves each campaign of conversion data and degrades the AI optimization. Splitting too broadly mixes business goals (lead-gen vs ecommerce vs catalog promotion) and gives the model conflicting signals.
Indicators that you should split: distinct conversion actions (sign-ups vs purchases) that should be optimized differently, distinct margin profiles where one product family is 60% margin and another is 20%, and geographic differences large enough that a single campaign's location-modifier handling becomes the limit.
What this looks like in Gapscout
Gapscout's Google Ads integration treats PMax as a first-class campaign type. The validator checks asset-group composition before upload (image count, ratio coverage, video presence) so campaigns ship at 'Good' or 'Excellent' strength from launch. The Google audit agent scores PMax campaigns against brand-cannibalization, search-themes coverage, and asset-group-strength signals — with recommendations attached to dollar impact.