How to test Google Ads AI Max: A/B test design and evaluation
AI Max is a layer within a Google Ads Search campaign that automates search term matching and ad assets. It is not a separate campaign type like Performance Max. Official setup guide
Define what will change, then compare spending and outcome quality through an A/B test. This guide covers preparation, setup considerations, and adoption decisions.
Turn a feature announcement into a business question
On August 20, 2026, Google announced testing and planning updates. Budget and ROI target comparisons across multiple Search campaigns were described as rolling out from September, alongside expanded testing with brand and location controls. Google announcement
Choose a question before combining features. “Use AI Max” is not an evaluation criterion. “Increase qualified inquiries under our current cost constraints” is a business question.
| Question | Suggested comparison |
|---|---|
| Can expanded delivery find more qualified prospects? | Align destinations and selling conditions, and record enabled features |
| Should we adopt the full feature package? | Evaluate the combination using costs and outcomes |
| What happens to profitability when spending increases? | Compare budget conditions separately from the initial adoption question |
Testing a complete package is valid. Its result does not isolate the contribution of search term expansion alone.
Prepare the evidence
Record:
- The campaign and current AI Max settings.
- The desired outcome and its measurement.
- Acceptable spending and acquisition cost, plus inquiry eligibility.
- Changed features and fixed conditions such as copy, landing pages, and bidding.
- Review timing, early stop conditions, and reasons to withhold judgment.
Resolve broken measurement or unclear conversion definitions first. Use the GA4 and Google Ads conversion reconciliation guide if the counts differ.
Set up an AI Max A/B test
In Google Ads Experiments, choose campaign features and settings, then AI Max for Search campaigns. The dedicated test splits traffic and budget within the existing campaign between control and treatment. Official experiment help
Multiple features are enabled initially. Check text customization, final URL expansion, and search term matching before saving. Brand settings added through the experiment apply to both groups, so record them as comparison conditions.
Restrictions include shared budgets and other running experiments. Check the same help page when a campaign is unavailable. Assess any operational impact before changing the existing budget structure to make it eligible.
Check destinations and advertising promises
When using final URL expansion, review possible destinations beyond your expected page. Check controls for recruitment pages, discontinued products, or pages where purchase is unavailable. Setup guide
Rewriting the landing page while expanding delivery changes the comparison. Minimize concurrent changes and record unavoidable ones. Google's experiment guide also recommends deciding hypotheses and evaluation metrics in advance.
Decide whether to adopt, revise, or wait
| Observation | Check before deciding |
|---|---|
| More conversions | Cost, unsuitable inquiries, and progression to sales or purchases |
| Lower acquisition cost | Consistent conversion definitions and changes in branded demand |
| More clicks only | Search terms, destinations, measurement, and conversion delay |
| No clear difference | Whether the report and data volume support a conclusion |
| Other changes occurred | How they affected each group and whether retesting is necessary |
Clicks and form submissions alone do not establish a business improvement. If sales take time, check whether outcomes have matured by the end of the experiment.
AI planning template
Review a plan to test Google Ads AI Max.
Business and intended customer:
Campaign and current relevant settings:
Question to answer:
AI Max features to enable:
Settings and landing pages to keep fixed:
Outcome definition and conversion delay:
Acceptable spending and acquisition cost:
Unsuitable inquiry conditions:
Planned review timing:
Return:
- Missing information before launch
- Differences and shared conditions between control and treatment
- Conditions for continuing, revising, stopping, or withholding judgment
- Evidence to review in the results
Do not invent improvement rates or required sample sizes.
Do not attribute a package-level result to one feature alone.
Start with one sentence defining the question and scope. Shoryaku AI chat can help discuss evaluation criteria using connected Google Ads data and your business conditions. Configure AI Max experiments and review their reports in Google Ads itself.