AI Max in Search Ads: What Businesses Should Test Before Scaling

AI Max is an attractive proposition. The platform can find new searches, adapt the message, and choose a landing page automatically. The promise is simple: broader reach, less manual work, and—ideally—more conversions.

Still, whenever Google Ads or Microsoft Advertising introduces another layer of automation, my first question is not “how quickly can we increase the budget?” It is: how much freedom are we giving the system, and how will we know whether it is using that freedom in the business’s interest?

I have seen campaigns that looked healthy inside the advertising dashboard while generating overly broad searches, low-value form submissions, or leads from areas the business could not serve. That is why I see AI Max as something worth testing, but not something to switch on and leave unattended.

What changed, in plain English

In August, Google announced new AI Max testing and planning tools. Advertisers will be able to test budget and ROI target changes across multiple Search campaigns while keeping brand and location controls active during the experiment.

The change matters for another reason as well. Google has said that, starting in September 2026, eligible campaigns using older features—including Dynamic Search Ads—will begin upgrading automatically to AI Max.

Microsoft Advertising is moving in the same direction. On August 27, 2026, the company announced that AI Max for Search was available, built around three main capabilities: broader search term matching, text customization, and final URL expansion.

In other words, this is no longer an isolated experiment on one platform. Search advertising is clearly moving toward more automation.

Why I would not start with a large budget

In a traditional campaign, you choose the keywords, write the ads, and decide which landing page the user sees. AI Max gives the platform more room to make decisions in all three areas.

That can be valuable when the account has reliable data, the website is well structured, and conversions are measured correctly. But the additional freedom also magnifies existing problems. If tracking is wrong, the AI will optimize more efficiently for the wrong outcome. If the website contains weak pages, final URL expansion can turn them into landing pages. If location targeting is too broad, the campaign may produce more leads the company cannot serve.

I would therefore treat AI Max as a business experiment, not as another box to tick in the platform.

What I would check before testing

1. Which searches trigger the campaign

I would start with the search terms report. I want to know whether the campaign is entering competitor searches, unrelated variations, or queries that look relevant at first glance but have no commercial intent.

Not every new click is progress. Sometimes automation discovers useful demand. At other times, it simply buys broader traffic. You only see the difference when you connect search terms to conversions and lead quality.

2. Where those people are located

For local and regional businesses, location is not a minor setting. A clinic in Iași, a repair company in Cluj, or a B2B provider serving only Romania gains nothing from cheap leads in areas it cannot cover.

Before the test, I would review presence-versus-interest settings, accepted cities or countries, and the actual geographic report. Once the campaign is live, I would keep returning to that report instead of looking only at total conversions.

3. Which pages AI Max is allowed to choose

Final URL expansion has real potential, but it is also one of the features I would watch closely. The system may find a page that fits the search better than the manually selected landing page. It may also route traffic to an outdated page, an informational article with no clear CTA, or a service the business no longer wants to prioritize.

Before giving it that freedom, I would clean up the site: current content, a clear offer, visible calls to action, working forms, and URL exclusions for pages that should never be used in ads.

4. What the account calls a conversion

In my experience, this is where the most expensive mistakes happen. If a click on a phone number, a visit to the contact page, and a completed form are treated as equal outcomes, the platform will pursue whichever action is easiest to generate—not necessarily the most valuable customer.

I would keep only genuinely meaningful actions as primary conversions. Where possible, I would also send qualified-lead or closed-sale data back to the platform. AI Max needs good signals, not simply more signals.

5. Whether the leads have real value

Cost per lead is useful, but it does not tell the whole story. A lower CPL can hide spam, unreachable contacts, requests for services the company does not provide, or prospects with no realistic budget.

During the test, I would track how many leads can be reached, how many fit the offer, how many progress to a sales conversation, and how many become customers. That is where you learn whether automation is helping the business or merely improving a number in the dashboard.

How I would structure the test

I would begin with a campaign, or a group of campaigns, that already has enough history to provide a useful baseline. Without that comparison, it is difficult to know whether AI Max improved performance or merely changed how the budget was distributed.

Before launch, I would write down a few practical boundaries:

the budget I am willing to invest in the test
a period during which I will avoid other major changes
the baseline cost and conversion rate
excluded terms, brands, and locations
the pages the system may use
the criteria for a qualified lead
the threshold at which I pause or adjust the test

I would not draw conclusions after two days, especially in a low-conversion account. But I would not let the campaign spend for weeks simply because the platform says it is “learning,” either. The test needs enough time to produce meaningful data and enough control to stop problems from becoming expensive.

When I would scale

I would increase the budget only if AI Max delivered results at least as strong as the control and the benefit was visible beyond the advertising dashboard.

The signals I would look for are new but relevant search terms, leads from the right locations, suitable landing pages, and stable or improved commercial quality. If the only outcome is more clicks and more weak form submissions, the campaign is not ready to scale.

AI Max does not remove strategic work. It moves it. We will probably spend less time adjusting every keyword manually and more time checking data, pages, rules, and real business outcomes.

That is the principle I would use for the test: let the AI find new opportunities, but clearly define the field it is allowed to play on.