Performance Max Best Practices for 2026: What Actually Works

Performance Max is Google's flagship campaign type and the one most operators get wrong. The pitch from Google is "give us your assets and let our AI do the rest." The reality is that PMax has more hidden controls than any campaign type Google offers, and operators who know them outperform operators who don't by 2-3x.

This guide covers Performance Max best practices for 2026 from running PMax across customer accounts spending millions per month combined. The tactics here are the ones we have seen move actual ROAS, not the obvious ones in Google's official help docs.

What is Performance Max

PMax is the most-used Google Ads campaign type for retailers in 2026 and the source of the most operator frustration because it hides which placement, audience, or query produced the conversion. Google added a Search Terms report for PMax in 2023 (limited visibility), brand exclusions in 2023, and search themes in 2024.

Account structure that works

Most operators run too few or too many PMax campaigns. The structure that performs:

One PMax per goal

If the business has an e-commerce goal and a lead-gen goal, run two PMax campaigns. Never one. Mixing optimization signals across goals confuses the algorithm.

One PMax per major margin tier

Combine high-margin products (60%+) and low-margin products (15%-) in the same PMax and the algorithm will favor low-margin SKUs because they convert at higher volume. Split into separate campaigns with different target ROAS values: a high-margin campaign at 3x target ROAS, a low-margin campaign at 4x target ROAS, both calibrated to the actual breakeven ROAS for that margin tier.

Brand vs non-brand separation

PMax cannibalizes branded search by default. The 2023 brand exclusion feature lets you exclude your own brand entirely (recommended) or split into a brand-only PMax and a non-brand PMax with separate budgets.

Audience signals matter more than people think

Google says audience signals are "just suggestions." That is technically true but misleading. Three audience signals consistently improve PMax performance across the customer accounts we audit:

Customer match lists

Upload your existing customers (Google Ads Customer Match, requires hashed email or phone). PMax uses them as a strong signal for prospecting and finds lookalikes far more effectively than letting Google define the audience cold. Customer Match minimum is 1,000 matched users; below that, the audience won't activate.

High-intent custom segments

Build a custom segment from search terms your converting customers actually use, then pass it as an audience signal. We see 15-30% ROAS lift from this in customer accounts. Google's Custom Segments support keyword lists, URLs, and apps as inputs.

Excluded audiences

Existing customers (when the goal is new acquisition) and frequent low-converters should be excluded explicitly. Google added the new-customer-acquisition goal to PMax in 2023, which automates this.

Asset group strategy

Asset groups are PMax's "ad group" equivalent. Most operators put everything in one asset group, which throws away the only segmentation lever PMax gives you.

One asset group per product category

If you sell shoes and jackets, create one asset group for shoes (with shoe-specific images, headlines, descriptions) and one for jackets. The same image-headline combinations should not run on both.

Use the maximum asset slots

PMax accepts:

• 15 headlines (30 char max each)
• 5 long headlines (90 char max each)
• 5 descriptions (90 char max each)
• 20 images (1200x1200 square, 1200x628 landscape, 1200x1500 portrait)
• 5 video assets (10-30 second optimal; auto-generated if you don't provide)
• 5 logos
• 1 site link extension minimum

Filling all slots is mandatory; partial assets get penalized in the auction.

Refresh assets every 6-8 weeks

PMax creative fatigues. Adding 3-5 new assets every 6-8 weeks keeps performance steady. Replacing the lowest-performing asset is enough; you do not need a full refresh. Asset performance scores (Best/Good/Low) appear in the Asset Group Performance report; replace Low-scored assets at the 6-week mark.

Brand exclusions and search themes (2024 additions)

Two controls Google added that change the game:

Brand exclusions (added 2023)

Account-level brand list. Add your own brand to it; this stops PMax from cannibalizing branded search. Add competitor brands you do not want to bid on. As of 2024, brand exclusion lists support up to 1,000 entries per account.

Search themes (added 2024)

A "hint" to Google about what queries you want PMax to compete on. Treat it like an exact-match hint: 5-15 specific themes that describe your top-converting search terms. Vague themes ("running shoes") are useless; specific ones ("women trail running shoes size 8") direct the algorithm. Search themes operate at the asset-group level; you can have different themes per asset group.

Smart Bidding thresholds and target ROAS tuning

PMax exclusively uses Smart Bidding (no manual bid option). The two strategies:

If you switch to Target ROAS below the 50/30d floor, the campaign starves volume. Stay on Maximize Conversions or Maximize Conversions with CPA cap until the threshold is hit, then switch.

For target ROAS tuning: start with the account's historical Shopping ROAS as the target. Adjust by 5-10% increments every 14 days based on volume and account-level ROAS. Setting target ROAS too high constrains volume; too low burns budget on weak conversions.

What we are tracking weekly

How Hyper helps

Most of the controls above require pulling data from multiple PMax reports daily and acting on it. Hyper's Google Ads agent runs the search term harvest, audience signal updates, asset score review, and brand exclusion checks hourly. Operators get a single weekly review instead of 7 hours of manual PMax surgery. Across 1,000+ customer accounts and 10M+ USD/month managed ad spend, Hyper customers running PMax typically see 15-25% ROAS lift in the first 60 days vs the manual baseline. The case study at /blog/ai-marketing-case-study covers the full setup we use.

Frequently asked questions