Google is changing the machinery behind paid search. Starting in September 2026, eligible campaigns using Automatically Created Assets and campaign-level broad match will begin moving to AI Max. Dynamic Search Ads will follow with automatic upgrades beginning in February 2027, according to Google's transition announcement.
The easy response is to treat this as another feature launch. Turn it on, accept the recommendations, and wait for the conversion column to improve.
That would miss the point. AI Max makes paid search more automated, but it also makes weak marketing harder to hide. If the offer is vague, the landing page is generic, the conversion signal is noisy, or the exclusions are poorly maintained, more query coverage won't rescue the campaign. It will just spread the confusion further.
The setting is not the strategy
Google says AI Max combines advertiser inputs with broader intent signals, search term matching, text customization, and final URL expansion. The company reports an average of 7% more conversions or conversion value at a similar CPA or ROAS for non-Retail advertisers using the full feature set compared with search term matching alone. That is Google's internal data, not a promise for every account.
The important part is what sits underneath the product language. AI Max is not replacing marketing judgment. It's making more decisions between the query and the ad, and between the ad and the page. That increases the value of the inputs you control.
Your offer. Your product pages. Your location rules. Your brand exclusions. Your conversion definitions. Your budget boundaries.
Those are not setup details. They're the campaign.
Google's own guidance on testing AI Max points advertisers toward experiments and performance planning. That is sensible. The mistake is running a test before deciding what a good result means.
Query expansion exposes weak offers
Broad matching and automated query discovery can find demand you didn't explicitly name. They can also find people who are technically related to your category but wrong for your business.
A tax law firm that sells high-value business counsel doesn't want every person searching for free tax advice. A local retailer doesn't want clicks from customers outside its delivery area. A B2B software company doesn't want students, job seekers, or people looking for a definition of the problem.
The platform can make a relevance guess. It cannot decide what your business can profitably serve.
That work starts with the offer. Who is this for? What expensive or frustrating problem does it solve? What proof makes the claim believable? What should a qualified visitor do next?
If those answers are unclear on the page, expanded targeting will multiply the number of unclear visitors. A campaign may report more conversions while the sales team sees weaker leads. A retailer may see cheaper traffic while store revenue stays flat. The dashboard can look busier as the business gets less certain.
This is why paid search still depends on the fundamentals covered in my post on landing page conversion strategy. More traffic is not a substitute for a page that makes the decision easy.
Landing pages become control surfaces
Final URL expansion is one of the most consequential parts of the change. Instead of sending every searcher to one carefully chosen page, the system can select a more relevant destination from the site.
That can be helpful when the site is well structured. It can be destructive when the site is a warehouse of half-finished pages, outdated offers, thin category copy, or competing calls to action.
Treat the site like a product catalog for machines and humans. Each important page should make its audience, use case, proof, action, and boundaries obvious. Old promotions should be removed or clearly retired. Location pages should describe real service areas. Product pages should not make claims the business cannot support.
A clean site gives automation better material to work with. It also gives a human visitor a reason to trust the click.
Before enabling broader automation, review the pages that could receive traffic. Look for pages with high impressions and poor engagement, pages with strong traffic but weak qualified actions, and pages that rank or advertise for terms outside the offer. The goal is not to make every page perfect. It's to stop sending automated demand into avoidable dead ends.
Measurement needs a stricter job
AI Max will make the old attribution argument even louder. A platform will show which searches, assets, and destinations it associates with conversions. That information can help with optimization, but it shouldn't become the only definition of performance.
Start with the decision you need to make. Are you deciding whether to increase spend, change the offer, open a new market, or stop a campaign? Each decision needs different evidence.
Then separate three layers of signal:
- Business outcomes, such as contribution margin, qualified pipeline, repeat purchases, or store revenue.
- Behavioral evidence, such as qualified forms, call quality, product engagement, and second-purchase rate.
- Platform outputs, such as impressions, clicks, modeled conversions, and asset combinations.
Platform outputs are useful. They are not automatically business outcomes.
My post-click measurement breakdown makes the broader case for giving attribution a smaller job. Use it to steer and diagnose. Use experiments, holdouts, matched markets, or geo tests when the answer would change the budget.
Test the system, not the headline
Google's reported average lift is interesting. It is not your baseline.
A useful test compares a defined control against a defined change. Keep the budget, geography, audience, conversion window, and business-quality criteria visible. Write down what would count as a win before the test begins.
For one account, that might mean a 10% increase in qualified leads without a decline in close rate. For another, it might mean stable contribution margin while query coverage expands. A lower CPA is not a win if the customers are worse. More conversions are not a win if the sales team spends twice as long disqualifying them.
Watch the search terms and destinations closely during the transition. Review what the system found, not just what it claimed to improve. Build a negative keyword and exclusion process that has an owner and a cadence. Automated campaigns still need human maintenance, especially when the platform is allowed to explore more of the market.
The operator still matters
The most useful paid search skill in this transition won't be memorizing every new toggle. It will be knowing where automation should stop.
Let the system search for useful patterns. Don't let it define the business. Give it a clear offer, a site with real structure, conversion events tied to value, and enough boundaries to avoid buying the wrong demand. Then judge the result against customer quality and money, not just the platform's preferred score.
A small business owner reviewing a campaign at midnight doesn't need another promise that AI will find hidden growth. They need to know whether the next dollar is buying a better customer, a useful conversation, or another line in a report.
AI Max may produce more reach. It may produce better results for some accounts. It may also expose businesses that have been using keyword controls to compensate for unclear positioning and weak conversion paths.
That is not a reason to avoid the change. It's a reason to prepare for it honestly.
The part that won't automate
Google can expand a query set, write a variation, and choose a destination. It can't decide whether the promise is worth making, whether the proof is credible, or whether the customer experience deserves the spend.
Those decisions still belong to the marketer. They always have.