How to test AI Max for Search without losing control of your queries
Google announced AI Max for Search on May 6, with a beta starting later this month. Here is what it bundles, what Google claims, and how to run a clean test.
On May 6, Google announced AI Max for Search campaigns, a one-click suite of targeting and creative features that Google says will roll out to all advertisers globally in beta starting later in May. It bundles three things that Search advertisers have been able to use separately: expanded query matching, generated ad text, and landing page expansion. The decision for you is not whether it is good. It is how to test it so the result means something.
What is in the bundle
- Search term matching. Expands your existing keywords using broad match and keywordless technology to reach queries you did not list. Google says the AI learns from your keywords, assets and URLs.
- Text customization. Formerly called automatically created assets, it generates headlines and descriptions from your landing page, ads and keywords.
- Final URL expansion. Sends users to the page Google judges most relevant on your site, which may not be the one you set.
Google also described new controls: brand controls at the campaign and ad group level to include or exclude specific brands, locations of interest at the ad group level (available only with AI Max), and a URL parameter that shows search terms across all match types. Reporting changes include search terms with headlines and URLs, asset reports, and the ability to remove generated assets and control URLs.
Read Google's performance claim carefully
Google says advertisers that activate AI Max typically see 14 percent more conversions or conversion value at a similar CPA or ROAS, and 27 percent for campaigns that rely mostly on exact and phrase keywords. The footnote matters: it is Google internal data from 2025, covering campaigns with more than 70 percent of conversions from exact or phrase match keywords, for non-Retail advertisers. That is a vendor-reported figure on a specific population. If your campaigns are already mostly broad match, it describes someone else.
The company's case studies are single-advertiser results. They show what can happen, not what is likely.
Run it as an experiment, not a switch
Search is often where a mobile or ecommerce business gets its highest-intent traffic, so the downside of a bad test is real. We would set it up this way once the beta reaches your account:
- Pick a campaign with stable volume. You need enough conversions in the test window to see a difference. A campaign with a handful of conversions a week cannot answer this.
- Use a split, not a before-and-after. Compare a version with AI Max against a control that runs at the same time on the same budget share. Seasonality and promotions will swamp a sequential comparison.
- Keep brand out of the first test. Brand queries convert well regardless and can flatter the result. Use the brand controls, or test on non-brand campaigns only.
- Turn features on deliberately. Because the three features come as a suite, test the suite first, then check whether you can isolate the effect of text customization or URL expansion afterward. If you rely on specific landing pages, such as a page per product category or a page that matches an offer, URL expansion is the feature to watch most closely.
- Set a decision rule in advance. For example, a minimum number of conversions per arm and a threshold for cost per conversion, written down before launch.
What to monitor
Look at where the extra conversions come from. If they come from net-new queries at acceptable cost, the product is doing what it claims. If they come from queries you already captured elsewhere, you may be paying twice for the same customer. Review the search terms report weekly, add negatives for clearly irrelevant matches, and read the generated headlines. Remove any that overstate your product, make claims you cannot support, or break your brand or legal guidelines.
Check the final URLs being used too. A user sent to the wrong page converts poorly, and that cost is easy to blame on traffic quality.
A note on scope: AI Max applies to Search campaigns. Google's app campaigns are a different campaign type, so this matters most if you also drive web or ecommerce conversions through Search.
The practical takeaway: wait for the beta to reach your account, test on one stable non-brand campaign with a concurrent control, and decide the success threshold before you start.