If you opened a Google Ads account in the past few days and found a Search campaign behaving differently from how you left it, you’re not imagining things. Google has started moving some existing campaigns into AI Max, and it’s doing so without asking for permission on a campaign-by-campaign basis. The stated goal is to complete the upgrade of eligible campaigns by the end of September. That means anyone managing Google budgets has only a few days to understand what’s happening, before October’s data, and therefore the entire fourth quarter, starts being evaluated on a different system from the one on which the original goals were built.
The point that interests me isn’t the usual debate about automation, whether it’s good or bad, whether it eliminates work or makes it better. That debate has been over for a while. The real issue is something else: with AI Max, your keyword list stops being the boundary of your campaign. And if you don’t define the new boundary yourself, Google will define it by reading your website.
Which Campaigns Are Being Migrated and Which Aren’t
The migration affects two types of campaigns. The first is campaigns that use automatically created assets, which Google now calls Text Customization: the system writes headlines and descriptions based on the landing page, existing ads, and keywords. The second is campaigns with broad match enabled at the campaign level. After the upgrade, both operate within AI Max with Text Customization and Search Term Matching enabled.
Dynamic Search Ads, on the other hand, are staying out for now. Google had also announced an automatic upgrade for them starting in September, but after advertiser pushback, it moved the date to February 2027. Accounts built around DSAs therefore have a few extra months, and it would be a mistake to waste them. This is the right time to clean up your page feeds and dynamic targets before the platform does it for you.
Search Term Matching: Your Website Becomes the Keyword List
This is where the real change happens. With Search Term Matching, keywords still exist and structure the campaign, but they’re no longer the only entry point. AI Max can use your assets and landing page content to target searches that weren’t included in your keyword list. It’s the logic of keywordless matching brought into traditional Search.
What does that mean in practice? A poorly written, generic page that talks about ten different services becomes a noisy signal. If the landing page for a beauty salon talks about laser treatments, facials, and a “free consultation,” the system has three possible directions in which to expand, and there’s no guarantee it will choose the one that generates the most margin. For years, we’ve said that landing pages exist to convert. Now they also tell Google what you want to appear for. That’s a new role, and most Italian SMB websites aren’t ready to support it.
Final URL Expansion: When Google Chooses the Page for You
The second piece is Final URL Expansion. If Google determines that another page on your domain is more relevant to a search, it can send the user there instead of the landing page you originally selected. On paper, that makes sense. In the reality of a local account, or in a regulated industry such as healthcare or finance, it can become a serious problem: a user ends up on your “About Us” page instead of the form, or on an SEO blog article instead of the page containing your offer. You pay the same for the click, but you don’t get the same conversion.
That’s why URL Inclusions and URL Exclusions exist, allowing you to decide which areas of your website can enter delivery. They’re not something to fix later. They’re the first thing you should configure, especially if your website has a blog, institutional pages, or sections that weren’t designed to receive paid traffic.
The Controls That Remain and the Ones That Really Matter
Google hasn’t taken the steering wheel away completely. Brand Inclusions and Brand Exclusions, location settings, and URL inclusions and exclusions remain available. Reporting is also becoming more detailed: in the search terms report, you can see whether a query was reached through broad match expansion or keywordless matching, and you can view the search, the headline shown, and the landing page together. This is useful because, for the first time, it allows you to think not about an individual keyword but about the combination that produced the result.
Google says that advertisers using Search Term Matching, Text Customization, and Final URL Expansion together achieve an average of around 7% more conversions or conversion value, at a similar CPA or ROAS, compared with those using Search Term Matching alone. That’s Google’s data, measured by Google, on Google’s platform. I wouldn’t treat it as a benchmark for your account, but rather as a statement of intent: the system is designed to work when you allow it to move all three levers together. The question is whether you have the tools to recognize when those levers are moving in the wrong direction.
The Real Risk Is Optimizing on an Incomplete Signal
There’s an issue I don’t see discussed much in the conversation around the migration, and it’s the one that concerns me most. AI Max expands coverage, meaning it looks for new searches, and decides which ones to pursue based on the conversions you feed it. If your tracking only counts completed forms, while calls, WhatsApp messages, and requests for directions end up in the “organic” bucket, the system is learning from only a fraction of reality. And the more freedom it has to explore, the more that error gets amplified.
Put differently: with the old structure, incomplete tracking caused you to make bad decisions. With AI Max, incomplete tracking causes the machine to make bad decisions, at a speed and scale that you can’t see. I’ve already written about how Google changed the rules around target CPA and target ROAS for budget-limited campaigns: the underlying principle is the same. The more the platform makes decisions on its own, the more the quality of the signal you give it becomes the only thing you truly control.
What to Do Before September Ends
The first step is to capture the current state of your account. Export CPA, ROAS, conversion volume, your main search terms, and the landing pages performing best, and save them. If the numbers change a month from now, you’ll need to be able to tell whether it was the migration or seasonality, and without a baseline, you’ll never know.
The second is to check the conversion actions each campaign is actually bidding toward. Not the ones you think are configured, but the ones that are actually configured. It’s more common than you might think to find secondary conversions promoted to primary, or, conversely, calls from ads that aren’t being counted.
The third is to review your exclusions: brand, URL, and location. If you have a blog, exclude it until you’ve consciously decided that it should receive paid traffic. If you operate in a specific geographic area, check that your location settings aren’t including people who merely “show interest” in the area instead of people who actually live there.
The fourth, the most boring and most important step, is to rewrite your landing pages with the understanding that they’re now also an input for matching. One page, one service, one intent. Everything else is noise that the machine interprets as permission.
AI Max Doesn’t Take Your Work Away, It Moves It
Anyone who thinks AI Max means less management is looking at the job the wrong way. The work doesn’t disappear. It moves from inside the campaign to outside it: from tweaking bids and match types to managing tracking, website architecture, assets, and the limits you give the system. It’s the same transition we experienced on Meta with Andromeda and Advantage+, where creative became the new targeting. On Google, the new targeting is your website, together with your conversion data.
So the question isn’t whether to accept or reject the migration. It’s a different one, and it’s worth asking now, before the Black Friday numbers arrive to act as the judge: do you know exactly what you’re teaching the algorithm, or are you finding out from the results?
