AI Max for Search campaigns: what it actually does, and how to keep it in check

AI Max arrived in beta during 2025 and left it in June 2026, when Google said hundreds of thousands of advertisers were already scaling Search campaigns with the feature set. Since then the question we get is rarely "what is it" and almost always "should I turn it on?". The short answer is: it depends how tidy your account is. The long answer requires understanding that AI Max is not one thing.

What AI Max actually is

AI Max is not a new campaign type. It is a switch inside existing Search campaigns that enables three distinct behaviours together:

1. Advanced query matching. The system matches your ads to searches your keywords — broad match included — would not have picked up, using your site and asset content as signal.
2. Text customisation. Headlines and descriptions can be adapted to the individual search, generating variants from your assets and landing pages.
3. Final URL expansion. The system can choose a landing page other than the one you set, when it judges another page a better answer to the search intent.

Those are three different delegations carrying three different risks. The first widens demand; the second touches tone of voice; the third touches the post-click experience, which is where your conversion rate lives. Knowing this matters, because each has its own controls — and because in a test you want to know which of the three is producing the effect you are measuring.

The controls you still have (and almost nobody configures)

The recurring criticism of AI Max — "it is a black box" — was more accurate a year ago. Through 2026 Google added a set of levers that fall into four families.

Brand exclusions versus negative keywords

This distinction is the most common source of error. If your goal is to keep ads off brand searches — your own, a competitor’s, or a reseller’s — the right tool is brand exclusions, not a negative list. Exclusions cover the brand, its common misspellings and related brands, which a hand-maintained list never manages to do completely.

Negative keywords remain the right tool for two jobs: brand safety (themes you never want to touch) and obviously irrelevant terms. Using them as your primary targeting lever inside AI Max is counterproductive: an aggressive negative list narrows the space in which the system can find value, and in some cases it collides with final URL expansion — the system is trying to route traffic to a page you are blocking by keyword.

When to enable it, and when to wait

Enable it if your account has stable conversion volume, clean tracking, enough site content for the system to understand what you sell, and you have already saturated the demand your keywords reach. In that scenario AI Max does exactly the job it was built for: finding adjacent demand you could not see.

Wait if you are in one of these situations:

• Conversions are few or erratic: the system has too little signal to learn from, and your test will be noise.
• Conversion value is unreliable — the same prerequisite we describe for Smart Bidding Exploration.
• Your site has many pages that must not receive paid traffic and you have not had time to configure exclusions.
• You operate in a regulated vertical where ad text cannot be freely rewritten.

Designing a test that means something

Most AI Max tests we see fail not because the feature does not work, but because the test is badly built. Three recurring mistakes: too short, under-budgeted, and a success metric chosen after the fact.

1. Use an experiment campaign, not a flat switch on your main campaign. You need a comparison, not a "before and after" contaminated by seasonality.
2. Give it 8-12 weeks. That is the window the industry reports as necessary for the system to settle: it is working across a far larger signal set, and the first weeks are pure exploration.
3. Define the metric first. If the goal is volume at flat CPA, say so; if it is efficiency at flat volume, say that. Switching metric halfway is the fastest route to believing a result that is not there.
4. Read the search terms weekly for the first four weeks. Not to intervene immediately, but to understand what kind of demand is being bought. Exclusions get added deliberately, not in bulk.
5. Watch where the traffic lands. Segment conversions by landing page: if URL expansion is routing traffic to worse-converting pages, you will see it here before it shows up in the totals.

Generated text: where a human still belongs

Text customisation is the part that worries brands most, and the worry is legitimate. The system builds variants from your assets and your site content: if the site carries an expired promotion, an old claim, or a phrasing legal would no longer approve, that material is potentially available to it.

Text guidelines exist precisely for this: to declare how the brand may be named, what cannot be promised, which register to use. Write them once, with whoever owns the brand, and revisit them whenever positioning changes. It is the kind of work nobody thinks of as "campaign management" — and in 2026 it is campaign management.

What to do now

1. List the pages that must never receive paid traffic and configure final URL expansion exclusions.
2. Replace brand-related negative keywords with the appropriate brand exclusions.
3. Write text guidelines together with whoever owns the brand.
4. Launch an experiment on a representative campaign, with a stated metric and at least eight weeks ahead of it.
5. If you still run Dynamic Search Ads, read the migration timeline toward AI Max: the deadline is closer than it looks.

AI Max is the direction Google is taking the Search network, so the useful question is no longer whether to enable it but with which boundaries. If you want to know which boundaries your account needs, the free audit starts from structure, tracking and landing pages — the three things that decide whether AI Max grows your business or buys you traffic you do not want. You can also see what we cover and the results we deliver.