Google Performance Max Asset Testing: How to Use the New Experiments Tool

Google Performance Max Asset Testing: How to Use the New Experiments Tool

Google launched asset experiments for Performance Max campaigns in June 2026, giving advertisers a structured way to test creative decisions before committing budget to a full rollout. The tool allows you to compare entire asset groups against each other, evaluate the impact of adding individual assets, test seasonal versus evergreen creative, and validate AI-generated assets from Google's Asset Studio. A centralized Experiments page consolidates asset tests and conversion lift studies in one location. This addresses the longest-standing Performance Max criticism: the system optimized automatically but made it difficult to understand which creative changes actually improved results.

Key takeaways

1. Google's Performance Max asset experiments launched in June 2026 and allow direct comparison of asset groups within a single campaign, with the winning variant declared based on a primary or secondary KPI.
2. Secondary KPI support enables testing against competing objectives simultaneously: for example, maximizing conversions while maintaining a target ROAS efficiency threshold.
3. The Experiments page consolidates asset experiments, conversion lift studies, and campaign-level experiments in one location, reducing the fragmentation of previous experiment workflows.
4. Asset Studio integration lets you test AI-generated creative directly against existing assets, with results measured in the same experiment framework.
5. API and MCC (manager account) support launched in the weeks following the main announcement, enabling agencies and large advertisers to run asset experiments at scale across multiple accounts.

Why Performance Max lacked asset testing before

Performance Max (PMax): Google's campaign type that runs across all Google inventory (Search, Shopping, YouTube, Display, Gmail, Discover) using AI-driven optimization. It combines all ad formats into a single campaign and automates creative rotation, audience targeting, and bidding.

The gap the asset experiments tool fills: PMax has always optimized creative rotation automatically, but the system's choices were opaque. An advertiser could see which assets received high, low, or learning asset performance labels, but could not run a controlled test to determine whether swapping a headline or adding a new video actually improved conversion rate. The only option was to make a change and observe overall campaign performance, which conflated creative changes with auction fluctuations, seasonal effects, and algorithm updates.

Asset experiments introduce a control group, isolating the creative change from other variables and producing a statistically meaningful result before you commit the full campaign budget to the new creative direction.

How the experiments tool works

The ICP problem this solves for performance marketing teams: decisions about creative refresh cycles, seasonal asset swaps, and AI-generated versus human-created asset performance have historically been made on instinct or week-over-week trend analysis. The experiments tool provides a hypothesis-testing framework that produces data before the decision is made.

Four experiment types are available:

Compare asset groups: Test a new asset group (new headlines, descriptions, images, and videos) against an existing asset group. This is the primary experiment type for creative direction testing. Traffic is split between the control and test asset group according to your experiment settings, and results are measured against your declared success metric.

Evaluate individual assets: Measure the incremental impact of adding a specific asset (one new video, one new headline variant, one new image) to an existing asset group. This is the surgical version: if you want to know whether a product-focused video lifts conversions compared to a lifestyle video, this experiment isolates that variable.

Test seasonal versus evergreen: Compare a time-limited seasonal asset set against your standard evergreen creative. This is particularly useful for peak season planning: run the experiment in the 4-6 weeks before peak, get data on whether the seasonal creative outperforms before committing to full seasonal budget.

Validate AI-generated assets: Test assets created through Google's Asset Studio against existing human-created assets using the same experimental framework. This provides a data-driven answer to the "should we use AI-generated creative" question rather than a subjective one.

Prooflytics surfaces creative lifecycle signals (Scaling, Mature, Fatiguing, Dead) per ad in the daily briefing. For PMax campaigns running asset experiments, the experiment results layer alongside creative performance signals to give a complete view of whether a creative change is warranted before and after the switch.

Prooflytics

Stop stitching platform exports together

Every channel in one brief — plus the memory of what each one actually drove.

Start free trialSee pricing

14 days free · no credit card

01. How to set up your first asset experiment

Step 1: Open the Experiments page. In Google Ads, navigate to Campaigns, then Experiments. All active and completed experiments, including conversion lift studies, are visible here.

Step 2: Create a new experiment. Click New experiment and select Asset experiment. Choose the Performance Max campaign you want to run the experiment on.

Step 3: Configure the control and test variants. Select your existing asset group as the control. Create or select a new asset group as the test variant. If you are testing individual assets, specify which asset to add or remove rather than swapping the entire group.

Step 4: Set your success metric. Choose your primary KPI (conversions, conversion value, or ROAS). If you need to balance competing objectives, add a secondary KPI. The secondary KPI support is specifically designed for cases where an advertiser wants to maximize conversions without allowing efficiency to fall below a threshold.

Step 5: Set duration and traffic split. Google recommends running experiments for 4-6 weeks to accumulate statistical significance. The traffic split (typically 50/50 for asset group comparisons) determines how quickly you reach significance; larger test audiences reach it faster but expose more budget to the test variant.

Step 6: Review and apply. When the experiment reaches significance, the Experiments page displays the result with a recommended action. Apply the winning variant to the campaign with one click.

02. What to test first

Not every asset change warrants a formal experiment. Prioritize experiments where the creative decision has meaningful budget exposure:

High-value test candidates:

• New product category landing page versus existing product mix in Shopping formats
• Video assets (30-second versus 15-second, voiceover versus text-on-screen)
• Headline framing (price-focused versus benefit-focused versus urgency-focused)
• Seasonal versus evergreen for the 4-6 weeks before your peak period

Lower-value test candidates:

• Minor headline word variations with similar meaning ("Shop now" versus "Buy now")
• Image color palette differences without substantive content change
• Assets that have already accumulated a Strong performance label with high volume

What to watch

• Experiment reaching significance before the 4-week minimum: if you see a declared winner in week 1, check that the traffic split was not too large. A rapid significance result on small traffic can be a false positive driven by algorithm variance rather than true creative lift.
• Secondary KPI being violated during an experiment: if the experiment maximizes conversions but your secondary KPI (target ROAS) is being missed, pause the experiment before it runs to completion and adjust the test variant.
• Learning period overlap: new asset groups in PMax enter a learning period when first activated. If the test variant is still in learning when the experiment ends, the comparison is not valid. Ensure the test variant had sufficient run time before the experiment start to exit learning.
• Asset Studio results significantly outperforming human-created: this is a signal to audit your existing creative process, not just celebrate the AI result. Google's AI-generated assets are optimized for its own ranking signals, which may differ from what drives long-term brand recognition or conversion quality.

Bottom line

• Google's Performance Max asset experiments end the era of making creative decisions based on overall campaign trend analysis; controlled tests now isolate creative variables.
• The four experiment types cover the most common testing scenarios: full asset group replacement, individual asset addition, seasonal versus evergreen, and AI-generated versus human-created.
• Secondary KPI support enables experiments that optimize for a primary goal without sacrificing an efficiency constraint, which is critical for ecommerce accounts balancing conversion volume against ROAS targets.
• Prioritize experiments on high-exposure assets (video, seasonal, landing page intent) where the decision has material budget impact.
• For teams tracking PMax creative performance in Prooflytics: the experiments tool produces the before/after data that complements the Scaling/Mature/Fatiguing/Dead lifecycle classification already visible in the briefing.
• You can read independent reviews of Prooflytics on G2 and compare it to alternatives in the marketing analytics category.