Google AI Max: What 23 Tests Say Before You Switch It On

Quick Answer

AI Max is a set of features applied to your existing Google Search campaigns, not a new campaign type. Switch it on deliberately, feature by feature, not as a single toggle. According to a March 2026 Search Engine Land analysis of 23 tests across 16 advertisers, campaigns using all three features together saw a 40% higher uplift in test success rates. The comparison group had opted into baseline search term matching alone. Per the same analysis, only about half of tested advertisers had enabled text customization, and 44% had enabled URL optimization. If your AI Max test disappointed, check which of the three you actually had running.

The Three Features People Treat As One

AI Max for Search campaigns bundles separate capabilities behind a single name. That naming is the source of most of the confusion.

• Search term matching broadens which queries your campaign can appear on, beyond your existing keywords.
• Text customization generates and adapts ad headlines to the individual query.
• URL optimization sends traffic to a more relevant page on your site than the one specified.

The first changes who sees your ad. The second changes what the ad says. The third changes where the click lands.

Switching on only the first gives you broader reach with your existing ad copy pointed at your existing landing page. That is the configuration most likely to produce a wider set of queries served by a less relevant message. It is also, according to the 2026 test data, the configuration a large share of advertisers were running when they judged the feature.

What The Test Data Actually Supports

Per the Search Engine Land analysis published in March 2026, across 23 tests run by 16 advertisers:

Two honest caveats before you act on that table.

First, 23 tests across 16 advertisers is a small sample. It is directional evidence from practitioners, not a controlled study, and Search Engine Land is owned by Semrush. Treat the direction as useful and the exact percentage as approximate.

Second, "uplift in test success rates" is a measure of how often a test succeeded, not how much revenue it produced. A feature set can win more often and still win small. Read it as evidence about which configuration to test, not as a forecast of your return.

The Adoption Gap Is The Interesting Finding

According to that analysis, the headline number is the 40%. The number that should change your behaviour is the adoption one.

Roughly half of tested advertisers had text customization switched off, and a majority had URL optimization switched off. So a large share of the AI Max opinions circulating in 2026 were formed on a partial implementation.

That reframes the usual complaint. "We tried AI Max and it brought irrelevant traffic" is a predictable outcome. You enabled broader matching without the two features that adapt the message and the destination to those broader queries. The traffic got wider and the response to it stayed fixed.

This is worth saying plainly because it cuts against the instinct most careful advertisers have, which is to change one variable at a time. That instinct is correct for testing a bid strategy. It is misleading here, because these three features are designed to compensate for each other. Testing the broadening feature alone is not a conservative test. It is a test of the least favourable configuration.

What Text Customization Actually Costs You

Now the part that rarely appears in coverage of this feature.

Text customization modifies your headlines. For a D2C brand selling straightforward products, that is usually an acceptable trade for relevance. For others it is not a small decision.

Consider financial services, healthcare, education, or any category where claims are regulated. An automatically generated headline is copy you did not write and did not approve, running under your brand name. If your legal or compliance team signs off ad copy, they need to know this feature exists before it is enabled.

The same applies to brands with a tightly controlled tone, and to any advertiser running in a market where a mistranslated or oddly phrased headline would read badly.

The tradeoff worth naming: the feature that the test data most clearly associates with better returns is also the one that gives away the most control. There is no version of this where you get the relevance benefit while keeping every headline exactly as written.

A decision rule. If your ad copy needs approval before it runs, do not enable text customization until you have run a sample of generated headlines past whoever gives that approval. If it does not, enable it as part of your first test rather than holding it back.

How To Test It Without Wasting A Quarter

Pick one campaign, not the account. Choose a campaign with enough conversion volume to produce a readable result, which usually means one that converts in double digits per fortnight rather than a handful.

Enable all three features together for that campaign. You are testing the configuration the data supports, not isolating variables.

Before you start, record the campaign name, the date enabled, and which of the three features are on. Add current cost per acquisition, return on ad spend, conversion volume, and the landing pages in use. Note any promotion running in the period.

Run it for at least four weeks. Automated bidding needs time to settle after a material change, and a fortnight of learning-period data will mislead you.

Then review search terms, not just the headline metrics. The question is whether the broader matching brought queries you would have chosen to bid on. If it brought volume that converts but at a cost you cannot sustain, that is a bidding problem, not a verdict on AI Max.

Your five-minute action today: open one Search campaign and check which of the three AI Max features are actually enabled. Most people are surprised.

Tips For Running AI Max Sensibly

• Check your negative keyword lists before enabling broader matching. Broadening reach against a thin negative list is how budgets leak.
• Review the search terms report weekly for the first month, then monthly. The first month is when you catch the query types you do not want.
• Keep a screenshot or export of your headlines before enabling text customization. You will want the before state when you review.
• Watch cost per acquisition by match type rather than in aggregate, since broader matching changes the mix underneath a stable-looking average.
• Do not enable this across the account at once. A single campaign gives you a control group for free.
• If URL optimization is on, check where it is actually sending traffic. A more relevant page by Google's reading is not always your best converting page.

What To Keep An Eye On

Watch conversion volume and cost per acquisition together, because broader matching can improve one while quietly worsening the other. Watch which landing pages receive traffic if URL optimization is enabled. Watch the generated headlines for claims your business cannot substantiate, which matters most in regulated categories. And watch for a fuller dataset, because 23 tests is a starting point and this feature set is changing quickly.

FAQs

Is AI Max the same as Performance Max?

No. Performance Max is a campaign type spanning multiple Google inventory surfaces. AI Max is a set of features applied to Search campaigns, and you keep the search campaign structure underneath.

Can I enable the features separately?

Yes, and that is exactly why results vary so widely. The March 2026 test data suggests the separate-adoption pattern explains much of the disappointment reported with the feature.

Will this work for a small budget?

The constraint is conversion volume rather than budget size. If a campaign produces only a few conversions a week, no test on it will give you a readable answer in a month, regardless of the feature.

Should I turn it off if results drop in week one?

Not on week one. A material change to a campaign triggers a bidding learning period. Judge it at four weeks unless spend is running badly ahead of plan, in which case pause on the budget problem rather than on the performance reading.

Where This Leaves You

The useful reading of this data is not that AI Max works. It is that most people testing it were testing a partial version, and the partial version is the one most likely to disappoint. Before you form a view, check what you have enabled. Before you enable text customization, check who needs to approve ad copy in your business.

If you would rather have that test designed and run properly, that is what our pay-per-click management work involves. Our guide to Google Ads management in 2026 covers the account structure this sits inside, and how rankings turn into leads and sales covers the measurement question underneath any paid test. For a worked example of paid media rebuilt around profit rather than volume, see our D2C retailer paid media case study.