Google is automatically upgrading certain Search campaigns to AI Max starting September 1. Igor Ivitskiy warns that Google's promised 14% conversion lift assumes your CPA target stays achievable, but his data shows target-based bidding often overshoots by more than 10%. Before the upgrade, document your economic objective, constraints, stopping conditions, and conversion tracking for each campaign.
Google starts upgrading search campaigns to AI Max on its own. Not asking, just upgrading. In the next few minutes, I'll show you which campaigns this hits, what actually changes inside them, and the 4 lines you should write down this week. Let me show you the list first. It's on the screen. And it comes from Google's own help pages. If a search campaign uses text customization, which used to be called automatically created assets, that campaign will be upgraded automatically. Same for campaigns using the campaign-level broad match settings. Dynamic Search Ads are a slower story. You won't be able to create new ones from September, and the migration of the ones you already have now lands in February 2027. So check 2 things in your account today: text customization and campaign-level broad match. Those are the 2 that move on their own. Now, what does the upgrade actually change? 3 things. And the name tells you none of them. First, search term matching. Your ads can show on queries that are nowhere in your keyword list. Second, text customization. The system writes new headlines and descriptions for the person searching. Third, final URL expansion. It picks which page on your site to send them to. So it changed what you match, what you say, and where you land. And here is what Google promises for that. It's on the next screen. In their words, advertisers who activate AI Max in search campaigns will typically see 14% more conversions Or conversion value. At a similar cost per acquisition or a similar return on ad spend. The first line under it is Google's own internal data from 2025 for non-retail advertisers. I have no reason to think that number is invented. Read the promise carefully though. It has a condition inside it: more conversions at a similar cost per acquisition. So the whole promise leans on the one thing: the target you typed into the campaign has to hold. Most owners assume it does. The thinking goes like this: fine, let it explore. My target cost per acquisition is the safety rail. If it goes and finds expensive nonsense, the bidding pulls it back. And there is a real argument under that. Query expansion does first demand a keyword list never ceases. I've watched hand-built keyword sets leak money for years. The human is not always the safer option. But look at this screen. Because that Rails has a measured hit rate, take 1,950 campaigns from my dataset. Target-based bidding landed within 10% of its own stated cost per acquisition in 34% of them. 2/3 did not. Another 36% went over target by more than a fifth. Of the 622 campaigns carrying a target return, only 27% met it. The median one delivered 0.86 of what it was asked for. Now, I want to be exact about what that is and is not. It is a measurement of target-based bidding in general. It is not a measurement of AI Max. And it is not proof that automation beats or loses to a human because I don't have those same campaigns run by hand to compare against. What it does say is narrowed and more useful. Your target is an input. It is not a promise. So read it as a corridor. Look at the next screen. You typed $40. In the data, the range runs from about 0.9 of your target to about 1.4. So the number your business case has to survive is closer to $56. If your economics only work at 40. Typing 40 does not protect you. That is the real job now, and it's on this last screen. 4 lines. Write them before the upgrade, not after. 1. The objective. The economic number you agreed to be judged on in your own margin, not the return the platform reports. 2. The constraints. Your budget ceiling, your brand rules, and the parts of the account the system may not touch. For this upgrade, that means 2 decisions: do you want brand terms swept into the expansion, and do you allow it to pick your landing page? 3. The stopping condition. This is the one almost nobody writes. It's the observation that ends the test. Here's one as a single sentence: after 50 conversions, If the cost per acquisition on the new expanded queries is above 1.4 times my target, I turn expansion off. That's it. One sentence written while you're calm. 4. The audit gates. Check that your conversion tracking is honest before you measure any of this. A bad signal fed to a fast system just gets you to the wrong answer quicker. That's the shift in the job. You don't collect AI features; you set the terms they run under. In 19 years of pulling accounts apart, the leaks I found almost never come from machine being fast. They come from nobody writing down what should stop it. So before activating AI Max, open your search campaigns. Check which ones use text customization or campaign-level broad match. Because those are the ones that move. Write your 4 lines for each of them. Then, after the upgrade, look at your search terms report in week 1, not week 6. You want to see what it started matching while the spend is still small. I put the full account checklist on my site and the long argument on LinkedIn. Both links are below. And if you want AI to be measurable rather than magic, subscribe. New Profit Forensics every week.
