Google Ads now allows A/B testing inside Performance Max asset groups. You can test whether new creatives actually outperform the current set.
Before this, you had to trust the algorithm. Upload assets, wait, watch aggregate metrics. No way to know what worked — the new headline, the image, or just a lucky day.
Why it matters
Performance Max is a black box. The algorithm mixes headlines, descriptions and images on its own, showing them wherever it sees fit. Little control, almost no data on individual elements.
A/B testing solves part of that. Take your current asset group, create variant B with changes, and run the experiment. Google splits the traffic, calculates results and shows whether the difference is statistically significant.
You create an experiment inside the PMax campaign. Pick an asset group, duplicate it with new creatives as the test version. The system distributes impressions between control and test automatically.
Google reports whether conversion rate changed, whether traffic got cheaper or more expensive, and which version delivered more goal completions. If the test group wins, you roll the changes out. If not, you revert or try something else.
What to test
Obvious candidates: headlines — direct offer versus a question, price in the headline or not. Images — product closeup or in use. Descriptions — benefit-led or feature-led.
Do not change everything at once. Swap text, images and CTA together and you will not know what worked. Change one element at a time.
Limits
Google has not disclosed sample size or test duration. The platform likely decides when there is enough data. That means low-budget campaigns may run tests for a long time.
Testing is available only inside one campaign. You cannot compare two different PMax campaigns through this tool, only variants within asset groups.
What it means
For arbitrage teams and media buyers, this is a step toward transparency. PMax is still autopilot, but now you can at least test hypotheses instead of feeding the algorithm creatives and hoping.
For offers with narrow audiences or expensive CPA, it matters more. Every change has to pay for itself. A/B testing lets you avoid burning budget on guesses.
The tool is new. How it behaves across verticals and how accurately it measures remains to be seen. But the direction is right: more control, less blind faith in the algorithm.
