AI Max Is Replacing the Old Google Ads Playbook

Google is not asking Search advertisers to adopt one more feature. It is changing what a Search campaign is.

Query expansion, generated text, and landing-page selection are being consolidated into an AI layer that sits across existing Search campaigns. The old operating model—keywords define the market, ads define the message, final URLs define the destination—is becoming less literal.

Keywords, copy, and landing pages still matter. They matter differently. They are now inputs and boundaries for a system that assembles the path between them.

Refusing the change because automation removes control is a bad answer. Enabling everything because Google says it can find more conversions is a worse one. The third option belongs to the operator: migrate deliberately, set the boundaries before the expansion, and make the algorithm earn the next dollar.

The Migration Is Real, but the Dates Are Not All the Same

Google announced on April 15, 2026 that AI Max for Search was moving out of beta. That announcement grouped Dynamic Search Ads, automatically created assets, and the campaign-level broad match setting into a single September 2026 automatic-upgrade timeline.

Google updated the post on June 11. The current schedule separates the features:

Automatically created assets—now called text customization—and the campaign-level broad match setting remain scheduled to begin automatic upgrades in September 2026.
The Dynamic Search Ads sunset and automatic upgrade were extended to begin in February 2027.
Google's current Help Center says advertisers can still add DSA ad groups. The cited official sources do not give an unambiguous creation cutoff, so confirm account notices before planning around one.

So "AI Max migration" is not one account-wide switch on one universal date. Eligibility, notices, available controls, and feature mapping vary by campaign. Confirm the current Help Center guidance and the notices inside each account before acting on either date.

The schedule has already moved once. Treat the dates as a change-management deadline, not as an argument for a rushed rollout.

What Is AI Max for Search?

Google's AI Max documentation is direct about this: AI Max is not a new campaign type. It is an optimization layer activated inside existing Search campaigns, and it has two feature groups.

Search term matching expands beyond the keyword list using broad match, asset-based signals, landing-page content, and keywordless technology. It turns on with AI Max at the campaign level and can be toggled off per ad group.

Asset optimization covers text customization and final URL expansion. Text customization generates headlines and descriptions from existing ads, keywords, assets, and site content. Final URL expansion can send a click to a different relevant page on your domain. Google requires text customization to be on for final URL expansion, so the generated copy can match the page the system chose.

The surrounding controls sit at different levels, and that detail decides how tight your test can actually be:

Control	Where it is set
Search term matching	On with AI Max at campaign, off per ad group
Text customization	Campaign
Final URL expansion	Campaign, and requires text customization
Brand inclusions	Campaign and ad group
Brand exclusions	Campaign
URL inclusions	Ad group
URL exclusions	Campaign
Locations of interest	Ad group

Reporting improved alongside the controls. Google's AI Max reporting lets you filter the search terms report for AI Max matches, shows expanded and landing-page matches in the keywords report, exposes Google-generated content in the asset reports, and lists final URL expansion selections in the landing pages report.

That is materially more inspectable than "let the machine decide." It is not the same as control. Google still decides the auction, the match, the creative combination, and—when enabled—the destination, inside the boundaries you supply.

So the operator's job moves upstream: clean inputs, explicit exclusions, verified conversion goals, and a decision framework that does not rest on Google's own attributed outcome. That is the same argument as treating the ad account as a scoreboard rather than the game, applied to a system that now writes part of the play.

Google's Performance Figures Are Vendor Claims

When AI Max entered beta in May 2025, Google reported that advertisers activating it typically saw 14% more conversions or conversion value at a similar CPA or ROAS, and 27% for campaigns still drawing more than 70% of conversions from exact and phrase keywords. In the 2026 migration announcement, Google reported an average 7% more conversions or conversion value at similar CPA or ROAS from the full feature suite compared with search term matching alone.

Those figures are useful evidence of product direction. They are also Google internal data for non-retail advertisers—not an independent benchmark, and not a forecast for your account.

An aggregate platform result leaves the expensive questions open:

Were the additional conversions qualified, incremental, and profitable?
Did the treatment capture existing branded or organic demand?
Did conversion lag or modeled attribution change the apparent result?
Which sites had enough accurate content to support generated text and URL expansion?
What happened to marginal CAC as spend increased?

Measure the migration on business quality and marginal economics, not on whether the Google Ads interface reports more conversions.

Do Not Migrate an Account That Cannot Define Success

AI Max expands whatever system already exists. That includes its errors.

Hold the migration experiment if any of these is true:

Form fills or other soft events control bidding while downstream lead quality is unknown. Fix the bidding event before the matching layer.
Primary and secondary conversion actions are misconfigured.
Brand traffic, existing customers, or low-margin products distort reported results.
The site carries outdated offers, thin pages, support content that should not receive paid traffic, or claims that cannot legally appear in an ad.
Tracking templates break when the final URL changes.
Sales, finance, and the ad account disagree about revenue. That disagreement is a data problem, not a reporting preference.
The campaign is materially budget-limited.

That last one is Google's own caution: AI Max will not be effective when a campaign is limited by budget. Google also warns that dynamic URLs can produce broken expanded landing pages when combined with tracking templates, and that pinning of responsive-search-ad assets is not respected once final URL expansion is on.

Those are not small technical details. A regulated claim pinned to a specific position, a legal disclaimer, or a deliberately controlled landing page can be a business requirement. If the requirement cannot survive the enabled feature set, constrain the feature or exclude the campaign.

Automation adoption is not a maturity badge. Sometimes "not yet" is the correct operating decision.

Establish a Baseline the Treatment Cannot Rewrite

Before enabling AI Max, freeze a baseline for at least one full conversion cycle—longer when volume or seasonality demands it.

Capture four layers:

Layer	Baseline measures
Delivery	Spend, impressions, clicks, query mix, match type, landing pages
Platform outcome	Conversions, conversion value, CPA, ROAS, lag-adjusted trend
Business quality	Qualified rate, opportunity rate, new-customer mix, returns or cancellations
Economics	Contribution margin, allowable CAC, payback, marginal return

Export search terms, negative keywords, brand settings, location settings, asset reports, landing pages, URL rules, and change history. Record which conversion actions are primary. List every campaign using DSA, text customization, or the campaign-level broad match setting so your inventory matches Google's two different timelines.

Then write the decision rule before you look at the treatment:

Adopt AI Max when it adds qualified conversion value at or above the marginal return floor, without violating query, brand, URL, compliance, or customer-quality guardrails.

Replace "qualified conversion value" and "marginal return floor" with the company's actual numbers. The ceiling comes from contribution margin, payback, and cash—not from a target ROAS. A test without a declared pass condition becomes a story-writing exercise.

How Do You Control AI Max Before Expanding?

Run this preflight audit on every test campaign:

Control	Question	Required action
Conversion	Is one economically meaningful goal driving bids?	Fix goals before testing
Query	Which terms must never trigger delivery?	Add negatives and brand exclusions
Brand	Which brands may be included or excluded?	Configure at the correct level
Geography	Does stated interest matter more than physical presence?	Review locations of interest
URL	Which pages may or may not receive traffic?	Build inclusions and exclusions
Copy	Which words, claims, and tones are restricted?	Add text guidelines and review site copy
Tracking	Can any selected URL preserve attribution parameters?	Test dynamic URLs and `{lpurl}` use
Compliance	Can generated assets serve safely?	Exclude constrained campaigns or disable features

URL exclusions deserve more attention than they usually get. Exclude login, support, careers, legal, investor, out-of-stock, obsolete, low-margin, and non-converting informational pages unless there is a deliberate reason to buy traffic to them.

URL inclusions create a narrower test surface. Use them to give one ad group a coherent family of destinations. Pointing the system at the whole domain is not learning; it is spending.

Tracking is the failure that hides longest. If a selected URL drops your parameters, the campaign keeps running while your CRM slowly stops recognizing where revenue came from—which is why server-side conversion infrastructure should be verified before expansion, not after a bad month.

Site copy is now targeting and creative infrastructure. Vague positioning gives the system vague inputs. Offers that conflict across pages get reproduced in ads.

Use an Experiment, Not a Before-and-After Comparison

Google offers a one-click AI Max experiment that splits an eligible campaign, holding a share of it with AI Max off as control. The default treatment activates search term matching and asset optimization; you can disable search term matching per ad group, disable asset optimization at campaign level, and keep URL inclusions and exclusions in play.

The one-click flow is unavailable when the campaign already uses text customization, targets the Display Network, runs a portfolio bidding strategy or shared budget, uses bidding exploration, or already has an active experiment. Custom experiments remain the fallback when you need a separate copied treatment.

Use the narrowest experiment that answers the decision.

Does query expansion add valuable demand? Test search term matching with asset optimization off.

Does dynamic copy and destination selection improve the result? Test asset optimization after query quality is understood.

Does the complete suite beat the controlled baseline? Test the combined configuration only after the site, goals, and exclusions pass the audit.

Changing everything at once tells you whether the bundle works. It cannot tell you why. That trade is acceptable for a low-risk campaign with strong economics. It is not acceptable when compliance, brand, or landing-page control is material.

Google's reporting guidance recommends waiting at least two weeks after enabling AI Max before making changes such as adding negative keywords. Treat that as a minimum stabilization cue, not a test duration. The test also has to cover enough conversion cycles, sales lag, and volume to mean something.

Judge Expansion by Marginal Economics

AI Max is built to find demand beyond the campaign's familiar territory. So the question is not whether total CPA held flat. It is whether the added volume was worth the added spend.

Calculate:

Marginal CAC = change in acquisition spend ÷ change in new qualified customers

Then compare marginal CAC with the allowable CAC derived from contribution margin and payback policy.

Say the control spends $50,000 and acquires 100 qualified new customers. The treatment spends $65,000 and acquires 120. Average CAC in the treatment is about $542. The marginal CAC for the 20 added customers is $750.

If allowable CAC is $600, the treatment's average looks healthy while the expansion itself is unprofitable. That gap between the average and the next dollar is the whole reason marginal CAC governs scale decisions.

Also inspect:

Brand versus non-brand query contribution
Net-new versus existing customer mix
Selected landing pages and their downstream quality
Generated text accuracy and offer consistency
Qualified and closed-won rate by query source
Contribution margin, not only assigned conversion value

Platform attribution tells you which interactions received credit. It does not establish incrementality. For a material budget expansion, validate the business effect with a holdout, geo test, or another causal design when feasible.

Define Rollback Conditions Before the Migration

Rollback is not an admission that AI Max failed forever. It is a control that preserves capital while the inputs get repaired.

Pause or narrow the treatment when any of these occurs:

Qualified marginal CAC exceeds the approved ceiling after the minimum decision window.
Query expansion enters prohibited or strategically irrelevant territory.
Final URL expansion sends meaningful spend to disallowed or low-quality pages.
Generated assets create inaccurate, noncompliant, or off-brand claims.
Tracking breaks, CRM match rates deteriorate, or conversion definitions change.
New-customer mix or contribution margin falls outside the agreed range.

Rollback is not one switch either. Google's setup guidance turns search term matching off in ad group settings, and text customization and final URL expansion off individually in campaign settings. Record the exact configuration before you change it—including which brand, URL, and location controls were active—so a partial rollback does not quietly become a different test.

When a guardrail fails, diagnose the layer. A bad destination needs URL exclusions. Weak query quality needs negatives or narrower search term matching. Bad lead quality needs a better bidding event. A regulated campaign may simply remain unsuitable for dynamic assets.

Run the Migration as a 90-Day Operating Program

The timeline should follow the business's conversion cycle, but a workable program has four stages.

Days 1–15: inventory and baseline. Map affected campaigns, export controls, validate conversion goals, audit the site, and set economic thresholds.

Days 16–30: preflight and experiment design. Add brand, query, copy, location, and URL guardrails. Test dynamic landing-page tracking. Select low-risk, representative campaigns.

Days 31–60: controlled test. Launch the eligible split, resist unnecessary edits, and inspect search terms, match sources, generated assets, selected URLs, and CRM quality. Log every exception.

Days 61–90: business validation and decision. Let downstream outcomes mature. Reconcile platform, CRM, and finance data. Adopt, revise, hold, or roll back each feature by campaign class.

The executive report should state what changed, which controls were active, what remains unknown, and what was decided. "AI Max increased conversions" is not an executive conclusion. "Non-brand qualified pipeline increased at a marginal CAC below our ceiling, with no URL or policy exceptions" is.

The New Playbook Is Controlled Delegation

The old Search playbook rewarded manual coverage: more keywords, more ad groups, more hand-selected paths. AI Max moves the leverage to the quality of the system around the campaign.

That system needs a clean conversion goal, a site the algorithm can safely interpret, economic thresholds finance believes, and controls somebody actually reviews. The machine can explore queries, assemble messages, and pick pages. It cannot decide what profit means, which claim creates legal risk, or when an attributed conversion is not incremental. The same asymmetry shows up wherever Google assembles the experience—including ads served inside AI answers, where you get even less reporting than this.

So the decision is not "turn on AI Max" or "protect manual control." It is:

Migrate the campaign classes that pass the preflight, test each material expansion against a stable control, and adopt only the features that produce qualified marginal value inside the company's boundaries.

If you need an operator to build the migration inventory, the experiment, and the economic decision model, apply to work with us. The deadline belongs to Google. The control system belongs to you.