Brand Controls in AI Max Experiments: Who Should Test Now

Brand controls inside AI Max experiments are the feature a specific group of Google Ads advertisers has been waiting for: the ones who declined to test AI Max for Search because a test could not carry their brand inclusions and exclusions into the treatment arm. On August 20, 2026, Google’s Ads and Commerce blog said that friction is gone. It said two other things in the same post, and one of them is not live yet.

The three items are easy to blur together because the post presents them as one expansion of AI Max testing and planning. But Google’s own sentence structure separates them. Multi-campaign A/B testing carries an explicit future qualifier. Brand and location controls inside experiments, and Performance Planner’s projected-impact view with one-click apply, carry none; both are written as things you can do now. Search Engine Land’s same-day write-up, published three minutes after Google’s, did not preserve that distinction, and a reader working from the recap alone would not learn that only one of the three carries a September date.

This is a decision post, not a recap. It lays out what each feature is, what Google’s text and help-center documentation actually support, where the one genuine documentation gap sits (location controls inside the experiment flow), and then sorts advertisers into the groups that should run the AI Max experiment they avoided, the groups that should wait for September, and the group that needs to query its migration-date fields before designing anything.

The primary source is a short post on Google’s Ads and Commerce blog, published August 20, 2026. Three consecutive paragraphs describe three features in two grammatical treatments: one carries an explicit future qualifier, and the other two are written in the present tense with none. That contrast, occurring inside one post, is the strongest evidence available about what is live and what is not.

The first paragraph introduces multi-campaign A/B testing, framed as building on the one-click experiments AI Max already offers: a way to test different budgets and ROI targets across multiple Search campaigns in a single test. It opens its second sentence with “Rolling out in September,” and gives no day within the month.

The second paragraph covers brand and location controls. The tense is present: new capabilities in AI Max experiments “now let you run tests” with those settings enabled. There is no rollout language, no beta label, and no date.

The third paragraph covers Performance Planner, and it uses the same construction as the second: the planner “now allows you to see” how bidding or budget changes may affect existing campaigns, and you can apply those suggestions in one click. Again, present tense, no qualifier.

Why lean so hard on grammar? Because Google applied the identical present-tense, no-qualifier treatment to Performance Planner, a feature whose help-center page describes one-click implementation as a working mechanic today, and applied a different, explicitly future treatment to the multi-campaign feature. When a vendor differentiates availability inside a single three-paragraph post, the differentiation is the information. Reading all three as one September wave, or all three as live, discards it.

The table below is the reconciled view. Each row pairs Google’s own wording from the August 20 post with what the help center documents, and ends with the reading we would use to make a decision. The location row is the one to read carefully: it sits in the “live” group because Google’s sentence puts it there, but its documentation column is honest about what was and was not found.

The brand-controls change only matters if you understand what an AI Max experiment is, because the mechanic is what makes a guardrail testable without contaminating the comparison. Per Google’s “About AI Max experiments” help page, an AI Max experiment diverts a percentage of an existing campaign’s traffic and budget to a treatment arm with the AI Max toggle on, against a control arm with it off. No campaign copy is created. That is the structural difference from legacy custom experiments, which duplicated the campaign.

Google lists three benefits of that design over custom experiments: faster results, because everything runs inside a single campaign; fewer setup errors, because both arms are updated together; and a shorter learning period. Those are vendor framings rather than measured outcomes, and Google does not document what produces them. The structural fact behind the claims is the one to hold on to: the test runs inside the existing campaign rather than in a copy of it.

The six blockers deserve a closer look before anyone schedules a test, because they decide who can even start. The help page states you cannot create an AI Max experiment through this flow if the campaign already has text customization (formerly Automatically Created Assets) enabled, targets the Display network, is on a Portfolio bidding strategy, uses Shared budgets, uses Bidding exploration, or already has an active experiment. Shared budgets and portfolio strategies are common in larger accounts, so the practical first step is an inventory: which campaigns are eligible for the experiment flow at all, and which need a structural change before a test is possible.

If you are newer to the product, our AI Max general-availability playbook covers the campaign-level settings this experiment toggles on and off; this post assumes that background and focuses on the testing decision.

This is the half of the claim that is confirmed beyond the blog post. Google’s announcement links its “capabilities” wording straight to the AI Max experiments help page, and that page documents brand settings as a working part of the current setup flow. Step 5 reads: “Configure Brand settings such as ‘Brand inclusions’ and ‘Brand exclusions’.” The page carries no rollout date, no beta label, and no coming-soon language anywhere. It describes a flow you can follow today.

Three details in that documentation are what make brand safety testable rather than merely present. First, brand inclusions and exclusions can be added during experiment setup “even if your original base campaign uses legacy brand controls or has no brand controls enabled.” You do not need to have migrated your base campaign to the newer brand controls before testing. Second, the settings you add apply symmetrically, which is the point of the next callout. Third, the end-of-experiment behavior is defined: if the experiment is not applied, all AI Max settings revert; if the base campaign had legacy brand controls before the test, it returns to those; if it had none, any brand controls added during the experiment are removed.

For advertisers whose only objection to AI Max was “I cannot risk my brand exclusions in a test,” this closes the objection. The guardrails are set once, applied to both arms, and either persist (if you apply the experiment) or unwind (if you do not). If your concern runs deeper than the exclusion list, into what AI Max is permitted to write in the ad itself, pair this with our guide to AI Max text guidelines for brand-safe Search ads, which covers the text-customization controls that sit alongside brand inclusions and exclusions.

Here is where honesty about sources matters more than a clean story. Google’s blog sentence pairs the two controls: “If you rely on specific brand or location controls, A/B testing just got simpler.” The grammar puts location controls in the live group. But the help page the blog links to documents only brand settings in the experiment setup flow. It does not mention location controls at all.

Location controls do exist as a real, documented AI Max feature. Google’s “How AI Max for Search campaigns works” help page describes locations of interest as an optional, ad-group-level setting that lets you “Reach specific customers based on their geographical intent even in keywordless matches,” with more than one location of interest selectable per ad group, in addition to campaign-level location targeting. That page also carries no rollout language. What it does not do is describe testing those locations inside an experiment, the way the experiments page describes brand settings at Step 5.

Two explanations are consistent with the evidence, and we cannot choose between them from primary sources. Google may not have published experiment-flow documentation for location controls yet. Or the location setting may surface identically to brand settings inside the experiment setup, and Google did not write a separate walkthrough. Either way, the correct posture for an advertiser is conditional, not confident.

The third feature is the easiest to act on because it is not an experiment at all. Performance Planner “now allows you to see how changes, like bidding or budget targets, may impact your existing campaign performance,” and “In one click, you can apply those suggested changes directly to your campaigns.” Both sentences are present tense. Google’s Performance Planner help page describes the same mechanic in its implementation section: apply suggested budget or bid adjustments for each campaign directly into live campaigns with a single click.

The safety net is worth knowing before anyone clicks. Changes made this way can be monitored and undone through the Bulk actions interface afterward. That turns a one-click change to a live campaign from a leap into a reversible action, which is the right frame for a planning tool: project, apply, watch, and roll back if the projection does not hold.

Two boundaries on the claim. Neither the blog post nor the help page says which recommendation types the one-click apply covers beyond bidding and budget targets, so do not extrapolate it to creative, targeting, or asset changes. And a March 9, 2026 note on the same help page concerns the discontinuation of Display and Video campaign planning support; it is unrelated to this feature and is not a dating signal for it.

Set that sentence next to the Performance Planner paragraph and the contrast is the whole story. One feature is described as a thing that will happen; the other as a thing you can do. Advertisers who want to see the projected effect of a budget change on an existing campaign do not need to wait for September. Advertisers who want to A/B test budgets and ROI targets across several campaigns in one experiment do.

Sort yourself by the objection or goal that kept you from testing AI Max, not by the headline. Four groups cover nearly every advertiser reading this.

Group 4 is the one with a clock on it, and the clock was set before this announcement. On April 15, 2026, Google announced that Automatically Created Assets and campaign-level broad-match Search campaigns would auto-upgrade to AI Max starting September 1, 2026, with a gradual rollout expected to conclude by the end of the month. Dynamic Search Ads were separately delayed to February 2027, and we covered that DSA delay and what to do with the reprieve on its own; it does not apply to ACA or broad match. The two Campaign fields added in Google Ads API v25.1 on August 19 are the API surface for that migration, and querying them is the one pull to make before designing any experiment on those campaigns. Google has not documented whether those fields carry a date ahead of the migration or only fill in once it has run, so read an empty value as unresolved rather than as “nothing scheduled.” In the same window, Google has said manual language targeting will be removed from Search and AI Max for Search in late September, so the September change list is longer than this announcement alone.

Everyone in every group can use Performance Planner today. It is independent of the experiment features, works on the campaign as it stands, and has an undo path. Treat it as the low-risk first move while the experiment decision settles.

The decision matrix tells you which group you are in. The sequence below tells you what to do first, because the steps depend on each other and the September clock runs regardless of whether you act.

1. Query the migration fields. For any campaign using ACA or campaign-level broad match, pull the per-campaign migration fields before anything else. If a date comes back and it falls in early September, that changes the calculus for a voluntary experiment that needs weeks to read; if the fields come back empty, treat that as unresolved rather than as reassurance, and re-query on a schedule.
2. Inventory experiment eligibility. Check each candidate campaign against the six setup blockers: text customization already on, Display network, Portfolio bidding, Shared budgets, Bidding exploration, or an active experiment. Shared budgets and portfolio strategies are the usual culprits in mature accounts.
3. Open the experiment setup and look at Step 5. Confirm the brand-settings step is present in your account, and note whether a location-controls step appears beside it. That one observation resolves the location question for your account better than any article can.
4. Set brand guardrails once, for both arms. Add inclusions and exclusions during setup so they apply symmetrically. Decide in advance whether you will apply the experiment if it wins, because that determines whether the controls persist or revert.
5. Run Performance Planner in parallel. Project the impact of any pending budget or bidding change on existing campaigns, apply what holds up, and keep Bulk actions open as the undo path.
6. Park multi-campaign tests until September. Note the feature in your roadmap with Google’s own wording, and do not promise a stakeholder a cross-campaign budget test before it ships.

The pattern underneath all six steps is the same one we apply across paid media engagements: separate what the platform says is live from what it says is coming, verify the live part against the help center and your own account UI, and let the deadline calendar, not the announcement calendar, set your priority. Where the test itself is the hard part, our analytics and measurement work starts with the experiment design, so the read-out is trustworthy before a single impression is diverted.