A few weeks ago, I wrote the news version of this for PPC Live: DSAs are done, here’s the timeline, here’s your real deadline. As I kept thinking and reading about the update, I figured a follow-up on the DSA-to-AI Max migration itself would be a good idea.
Flag: Google has already moved the date once since that PPC Live piece went up: the DSA sunset and auto-upgrade now begin in February 2027, though campaigns using automatically created assets and the campaign-level broad match setting still auto-upgrade in September 2026.
Start with what the table leaves out
When Google announced the sunset, it published a comparison table next to it.
If you read the table straight, AI Max wins every row: better targeting, smarter creatives, more granular controls, deeper reporting, simpler workflows. However, if you’ve been following me for a while, you know I go through every point and ask: “What’s truly in it for the user?” or “What’s in it for Google?”
The Controls row
Let’s start with the Controls row, because it’s the most revealing. DSA gets “Minimal.” AI Max gets “Granular.” And, yes, both labels are true. But neither one tells you what matters here: having granular controls only protects you if they’re set before the system starts learning. Brand controls, text guidelines, URL rules, and negative keyword lists built categorically by intent rather than scraped reactively from the search terms report; each of those has to exist early on. The system doesn’t wait for you to finish setup before it starts deciding what “job well done” looks like.
For example, the February 2027 automatic migration will port your legacy DSA settings into AI Max and switch on all three features (search term matching, text customization, and final URL expansion) by default. That’s Google being genuinely helpful about continuity, and I’m not even saying that sarcastically. But if the “granular” controls were the whole point, the default wouldn’t be “mirror what you already had.” The controls would have to exist prior to. My recommendation follows what my peers have said: use the migration tools, just don’t default to them. Oversee the entire process.
The Targeting row
The Targeting row plays out the same way. The way it’s framed is that AI Max moves past landing page content to real-time intent signals, matching on what it thinks the user wants rather than what they typed. In the right account, that’s a real upgrade, but what you give up is legibility. DSA’s matching was limited, but it was visible enough that you could form a decent guess about why a query pulled your ad. AI Max matches broader, reads more context, and shows you far less about how it got there. The upgrade is more reach, but with less line of sight.
The reporting problem
I also want to think about this on the reporting side. When the matching logic gets harder to read, the search terms report stops being a live window and more of a rear-view mirror. By the time a pattern is obvious enough to act on, the system will likely already have retained the information on whatever it was seeing before you noticed. Brad Geddes documented the practical version of this: AI Max treats all your keywords as broad match regardless of the match type you set, and it takes credit for conversions your existing exact and phrase terms would have won anyway. So the incrementality numbers in your dashboard are more of a claim than a fact. As always, verify the numbers you see in the Google Ads dashboard against your backend before you act on them.
The issue most are having is that DSA was predictable enough that most of us carried a working model of it in our heads. AI Max gives you more capability but less visibility at the same time, which means it can find performance issues faster than you can audit them. That gap, between what the dashboard reports and what the system has actually learned, is the same one that opens up in Smart Bidding when conversion quality deteriorates while volume holds steady. I’ve cleaned up that mess enough times to know what it looks like early on.
The $1,700 lesson
Six days into an AI Max campaign on an emergency repair account, $1,700 had cleared, and exactly one conversion had come in. The system had expanded matching into queries like “how to buy a [product]” and “[product] how-to”; people shopping for or researching an appliance purchase. AI Max found semantically adjacent searches, assigned them relevance, and bid confidently on them; there was no way for it to know that someone buying the product and someone who’s just had theirs die are entirely different customers. This is simply the system functioning as designed.
What it was missing was any signal about the difference between purchase intent and repair intent (which is, quite literally, the entire premise of the account). That distinction doesn’t live anywhere the system can read unless you put it there. So it spent $1,700 discovering something that could have been handed to it before the first bid went out.
Categorical negatives do different work at launch than they do mid-campaign. At launch, they’re the brief you hand the system before it starts exploring; mid-campaign, they’re the cleanup after the search terms report tells you where it already went. For a repair account, that pre-launch list needs to cover purchase modifiers, how-to queries, and product discovery searches; the whole category of queries where someone’s relationship with the appliance is ownership rather than repair. Build that before the campaign goes live, and the system explores within a narrower lane from the start.
The system can’t distinguish “semantically adjacent” from “commercially relevant.” Categorical negatives are how you make that distinction legible before it goes looking with your budget.
Two phases, and the risk is in the first
The migration checklist has two phases, and almost all the risk lies in the first. The conversion audit, the categorical negative build, the text guidelines, the brand controls, and the URL exclusions from your DSA campaigns that don’t carry over on their own should all be backed up and implemented before the AI Max campaigns go live.
One specific thing that trips people up: legacy URL rules like “page title contains” and “page content contains” go read-only on upgrade, so if you relied on them to keep ads off out-of-stock products or empty category pages, rebuild those as URL exclusions or a page feed before you migrate. Google’s migration tools handle the structural move well; what they don’t do is check what you’re feeding the system, and they can’t tell you whether the signals you rewarded in DSA are the ones you actually want AI Max chasing at greater speed and scale.
Before you pause anything, export your DSA baseline while it’s still live: search terms, landing pages, conversion paths. Once the old campaign is off, you lose the reference, and you’ll want the before picture to judge whether AI Max actually replaced what DSA did.
Migrate early, while you can still compare
Google recommends upgrading now rather than waiting for the auto-upgrade, so your setup aligns with your business from day one. I agree. Migrate early because it gives you time to observe how the system handles your specific signals while DSA is still running alongside it as a reference point. Do it when you can still compare, not when you’re locked in and guessing.
Where expert testing comes in
Andrew Lolk has been running these migrations hands-on, and our thinking lines up on the fundamentals: treat AI Max with DSA-grade discipline, exclude your brand terms and junk URLs up front, and remember it’s three things at once (keyword expansion, text customization, and final URL expansion), not a straight swap.
His testing sharpens the “keep DSA as a reference” point above. When he enabled AI Max inside a campaign with an active DSA, almost nothing happened: 72 clicks out of more than 3,000. AI Max only really took over once the DSA campaign was paused. So running them side by side in the same campaign won’t teach you much; the new system stays quiet until the old one is off. The clean version: stand up an isolated AI Max campaign as the direct replacement for the DSA you’re pausing, rather than flipping AI Max on inside a campaign that already works and reading the quiet numbers as proof it’s fine.
The efficiency is a coin flip
AI Max earns its keep when the inputs are clean, the conversion tracking is tight, and there’s enough data for the system to learn from without inventing patterns. Smarter Ecommerce looked at morwith a median 16% higher CPA and a median ROAS change of roughly zero, with individual outcomes rangingedian 16% higher CPA and a median ROAS change of roughly zero, with individual outcomes ranging from 42% above baseline to 35% below it. The volume is there, but the efficiency is close to a coin flip… which is the whole argument in one data point: the inputs decide the result, not the tool.
Make the instructions right
The table Google published is a fair summary of what each tool offers; it just can’t tell you whether your account is set up to take what AI Max gives without handing it enough rope to go find performance you’d never have signed off on. That audit is yours. February is the hard deadline; the window between now and then is when the outcome is actually decided.
The system will follow your instructions, so please make sure they are right before you hand it more power to follow them.