Performance Max asset testing is a Google Ads experiment that splits the traffic of one campaign between two sets of assets, so you can prove whether new creative improves results before you replace anything. You set it up on the Experiments page, test one asset group at a time, and run it for at least 4 to 6 weeks.
Performance Max is a goal-based Google Ads campaign type that serves ads across Search, YouTube, Display, Discover, Gmail and Maps from one campaign, according to Google's Performance Max overview. An asset group is the set of headlines, descriptions, images, videos and logos inside a Performance Max campaign that Google combines into ads. Testing those assets is a form of ad creative testing, which is why this guide sits in the A/B testing hub. It covers how the test splits traffic, which questions it answers, when a result deserves your trust and which mistakes waste a test.
How does a Performance Max asset test split traffic between assets?
A Performance Max asset test splits the traffic of one asset group between a control arm and a treatment arm inside the same campaign. Control assets serve only to the control arm, treatment assets only to the treatment arm, and the common assets you leave out of both groups keep serving to 100% of traffic.
The control arm is the share of traffic that sees your existing assets (assets A), and the treatment arm is the share that sees other existing or newly uploaded assets (assets B), per Google's A/B testing assets page. Common assets are the assets you place in neither group. Both groups sit in one asset group, so both count toward the asset group limits.
Performance Max experiments split traffic, not budget, according to the Experiments FAQs. You choose the percentages during setup, the split cannot be changed afterwards, and when it is not 50/50 Google scales the reported results so both arms stay comparable. There is no separate fee, because the test runs inside your existing campaign and budget.
Google's asset testing page adds that a traffic split within a single campaign can reduce the learning period. While the test runs, the tested asset group is locked in view-only mode, so you cannot edit, add or remove its assets, and new treatment assets go through policy review before they serve.
Which creative questions can Performance Max asset testing answer?
Performance Max asset testing answers one question: does a different set of assets change results for the same asset group, with everything else held equal? It fits questions such as new versus existing creative, video versus no video, or a feed-only campaign versus one with added assets. It does not answer bidding, budget or campaign structure questions.
Google's help pages describe three experiment setups as of October 2026:
- Any assets (A/B testing assets): a control group of existing assets against a treatment group of other existing or newly uploaded assets.
- Assets for retail campaigns: a product feed-only campaign as control against the same campaign with added text, image and video assets, per Google's asset testing page.
- Video: the control arm serves without video, with both uploaded and auto-generated videos suppressed, while the treatment arm includes your videos, per the asset testing page.
Search Engine Land lists typical tests such as user-generated content against polished brand creative and different messaging or image styles. Other questions need another tool. Asset tests are one type of what Google calls optimization experiments, next to uplift and upgrade experiments that compare whole campaigns.
| Criterion | Asset A/B test | Campaign level experiment | Asset report |
|---|---|---|---|
| Question it answers | Does asset set B beat asset set A in one asset group? | Does a campaign type or setting change lift results? | How did each asset perform alongside the others? |
| Method | Traffic split inside one campaign | Control and treatment campaigns or arms | Observational, no control group |
| Result | Difference with confidence interval | Difference with confidence interval | Asset-level metrics, ratios are directional |
| Can you edit assets meanwhile? | No, the asset group is view-only | Changes are possible but not recommended | Yes |
| Best use | Deciding to replace or add creative | Uplift, upgrade or feature tests | Finding weak or tired assets to test next |
When is a Performance Max asset test result worth applying?
A Performance Max asset test result is worth applying when the test ran at least 4 to 6 weeks plus one to two conversion cycles, the scorecard marks your bidding metric with a blue asterisk, and its confidence interval sits entirely above zero. Reaching that point depends on access, a clean setup, enough time and a careful reading.
Check that your account can run the test
Google's help page About Performance Max optimization experiments: A/B testing assets (Beta) still carries the Beta label on October 4, 2026. Search Engine Land and PPC Land reported in January 2026 that the beta expanded beyond retail to all Performance Max campaigns, and Search Engine Land wrote on October 2, 2026 that the tests are rolling out, without a timeline. Google has not announced general availability.
To check access, open Experiments, start a new experiment and see whether "Any assets" appears as an experiment type for Performance Max. Then clear the blockers Google lists for failed experiment creation: a shared budget, a portfolio bid strategy, Smart Bidding Exploration, another experiment on the same campaign in the same dates, unsupported features, and an asset group above the 15-asset limit the error message names.
Set up one hypothesis per test
- Open Experiments. Go to Experiments in the Campaigns menu and select the plus button under the "All experiments" tab.
- Choose the variable. Under "What do you want to test?" select Assets, then "Assets provided by you" under "Choose a variable to test", then "Performance Max" as the campaign type.
- Pick the experiment type. Select "Any assets" for an A/B test of two asset sets, or "Assets for retail campaigns" or "Video" for the feed-only and video tests.
- Select the campaign and asset group. Use "Select campaign" to choose the Performance Max campaign and the asset group you want to test.
- Fill both arms. Review existing assets in the "Control arm" card, then select or upload the assets for the "Treatment arm" card, either as additions or as a full switch.
- Set the traffic split. Enter the percentage for the control and treatment arms.
- Name and schedule. Fill in the "Experiment name" field, keep or adjust the end date and select Schedule.
Dates differ by experiment type. The A/B testing page says the start date is set to the following day and the end date comes from the Experiment Guidance System, which calculates the duration needed for statistically significant results. The feed-only and video instructions say you can currently only choose today as the start date.
Give the test enough time and traffic
The Experiments FAQs recommend at least 4 to 6 weeks and one to two conversion cycles, and they discard the first 7 days as ramp-up: an experiment from December 1 to December 31 only reports data from December 8 to 31. The help page on monitoring experiments names the usual reasons for a result that is not statistically significant: the test has not run long enough, the campaign gets too little traffic, the traffic split is too small, or the change made no real difference.
The trust check turns those rules into one decision aid built from Google's documentation. It is not a Google feature.
| Condition | What to check | If it fails |
|---|---|---|
| Run time | At least 4 to 6 weeks and one to two conversion cycles after the 7-day ramp-up | Extend the end date or wait |
| Significance | The scorecard shows a blue asterisk at the confidence level you chose | Treat it as no clear winner |
| Direction | The confidence interval for your main metric sits entirely above or below zero | No decision; the arms may be equal |
| Clean test | No setting changes to text customization, Final URL expansion or video enhancements during the test | Restart the test |
| Traffic | Neither arm got a very small share; Google names a small split as a cause of weak results | Rerun with a split closer to 50/50 |
| Main metric | You judge on conversions or conversion value, matching the bid strategy, not on CTR | Recheck on the bidding metric |
A test that passes every row is a reasonable basis for applying the treatment. A test that fails one row is a lead for the next test, not a verdict.
Read the scorecard and apply the winner
The scorecard in the experiment report shows each metric's difference between arms, a confidence interval that defaults to 80% and can be changed, and a blue asterisk when a result is statistically significant, per the monitoring guide. The Results column names the control or treatment arm as the winner, or "No clear winner".
Worked example (hypothetical numbers). An asset group on Target ROAS tests new lifestyle images against existing product images on a 50/50 split. After 6 weeks the scorecard shows conversion value at +9% for the treatment with an interval of +2% to +16%, marked with a blue asterisk. The whole interval sits above zero, so the gain is likely real, though it could be as small as 2%. If the interval had been minus 3% to +21%, the same +9% would mean nothing yet.
Selecting Apply experiment gives two choices on the A/B testing assets page:
- "Add treatment assets to campaign": adds the B set to the asset group.
- "Keep control assets in campaign": selected by default, so the A set stays. Uncheck it to remove the A set and serve only the winners plus the common assets.
End experiment stops the test without changes and discards any new treatment assets. Conversions in the report follow your conversion settings and attribution model, so a change in either during the test muddies the comparison.
Common mistakes with Performance Max asset tests
Most failed Performance Max asset tests come from the setup, not from the creative: too many ideas in one treatment, too little traffic, a test cut short, or a winner picked from the asset report instead of the experiment. Each mistake below leaves you with a result you cannot act on.
- Testing several ideas at once. New headlines, new images and a new video in one treatment arm tell you the set won, not which part. Change one creative idea per test.
- Ending the test after two weeks. The first 7 days are discarded and Google recommends 4 to 6 weeks, so an early stop mostly measures ramp-up. Let it reach the end date the Experiment Guidance System suggests.
- Choosing a low-volume asset group. Google names too little traffic as a main reason for results that are not significant. Test the asset group with the most conversions first.
- Using the asset report as the verdict. Google calls asset-level ratios such as CTR and ROAS directional only. Use the report to pick the next test and the experiment to decide.
- Blaming creative for a landing page problem. If both arms convert poorly, the page is the bottleneck. Work on conversion rate optimization and check whether your Google Ads cost per click leaves room for the test.
