Audit Google Ads Before the September AI Max Upgrade

Current Google Ads documentation schedules an automatic upgrade to AI Max, starting in September 2026, for Search campaigns using the campaign-level broad-match setting. This is a future change rather than a completed migration, and it does not apply merely because a campaign contains individual broad-match keywords.

The enduring rule of advanced Google Ads management is unchanged: automation performs against the objectives, data and boundaries supplied to it. What is changing is the manager’s control surface. As query expansion, text customization and landing-page selection become more automated, conversion definitions, exclusions and experiment design carry more weight than routine keyword-level bid edits.

Identify the campaigns exposed to the upgrade

Start with a campaign inventory, not an account-wide assumption. Separate campaigns using the campaign-level broad-match setting from those that simply contain broad-match keywords, then record each exposed campaign’s bidding strategy, conversion goals, brand settings, negatives, landing-page configuration and budget status.

The campaign-level setting does more than standardize keyword syntax. It converts existing phrase- and exact-match keywords to broad match, makes subsequently added keywords broad by default and changes prioritization for identical queries. It is available only with conversion- or conversion-value-based Smart Bidding.

Preserve a baseline before changing the configuration. Capture cost, conversions, conversion value, cost per acquisition and return on ad spend at campaign and ad-group level, then retain the associated search-term, conversion-action and landing-page detail. Comparisons become unreliable if the conversion definition, attribution configuration or offer changes midway through the evaluation period.

Give automated bidding an economic objective

Automated bidding can optimize only toward the actions and values passed to it. Ecommerce accounts should verify purchase values, currency handling and duplicate-transaction controls. Lead-generation accounts need to distinguish commercially meaningful leads from low-intent submissions and, where the sales process permits, return later funnel outcomes to the advertising platform.

Each campaign should have a clear primary objective. Conversion-value bidding can distinguish between orders when reported values are dependable; conversion-volume bidding may be more suitable when qualified actions have broadly similar economic value. Treating page views, unqualified forms, phone calls and completed sales as equally important primary conversions gives the system a confused target.

Target CPA and target ROAS settings also require defensible economics. An acquisition target should reflect close rates, contribution margin and the value of the measured action. A revenue-based return target should be assessed alongside margin, fulfillment costs and any carefully supported repeat-purchase assumptions.

Move from keyword micromanagement to campaign boundaries

Google’s AI Max specification defines the product as an optimization layer within an existing Search campaign, not a separate campaign type. Search-term matching can draw on broad-match, asset-based and landing-page-based technology, while available controls include brand settings, locations of interest and URL inclusions or exclusions.

The manager’s task therefore shifts from predicting every useful query to defining where expansion is unacceptable. Brand inclusions can constrain activity that must remain connected to specified brands, while brand exclusions can block unwanted associations. Location-of-interest settings merit particular scrutiny when the place expressed in the query matters more than the searcher’s physical location.

Negative keywords still provide a direct boundary for irrelevant intent, prohibited categories and products or services the business does not offer. Review the list as a system: an overly broad negative can suppress valuable demand, while overlapping exclusions can make delivery difficult to diagnose.

Search-term analysis should connect intent with the complete customer journey. Ask whether the query is commercially relevant, whether the served message makes a supportable promise and whether the selected destination fulfills it. A costly term without a conversion is not automatically unsuitable; the sample may be limited, the destination may be weak or the recorded conversion may arrive later.

Audit generated text and landing pages together

AI Max can activate text customization and final URL expansion, although either can be switched off in campaign settings. Final URL expansion depends on text customization, and pinned responsive-search-ad assets may not be used when the system selects a different URL. Businesses with mandatory wording should consequently inspect the served message and destination together rather than assume a pinned asset will accompany every landing page.

The website itself becomes part of campaign governance. Exclude obsolete, noncommercial or legally unsuitable pages, as well as destinations that cannot complete the intended conversion. Product availability, prices, geographic eligibility and advertising claims should remain consistent on any page that automation may select or use to derive text.

Tracking templates also require attention because dynamic destinations can fail when URL parameters are incompatible. Verify that the template passes the selected landing page correctly and test representative expanded URLs for errors before relying on the feature.

Reporting should be reviewed at the query-message-destination level. AI Max reporting can associate search terms with headlines, URLs and match sources, making it possible to see whether apparently relevant traffic received an unsuitable claim or destination. That combination is now a more useful diagnostic unit than the keyword alone.

Test one strategic change at a time

Google’s Experiments documentation describes a dedicated AI Max test that splits an existing campaign between control and trial traffic; for results that remain undetermined, it recommends allowing at least four to six weeks before evaluating whether more time or budget is needed. Custom experiments remain available for changes involving Smart Bidding, match types, landing pages and ad groups.

An experiment is useful only when it isolates a decision. Turning on AI Max while replacing the conversion action, restructuring landing pages and rewriting the offer would make any change in performance difficult to attribute.

Write a hypothesis tied to an economic outcome, such as increasing qualified conversion value without exceeding the allowable acquisition cost.
Select the primary decision metric before launch and define guardrails for lead quality, brand traffic, geographic fit and landing-page compliance.
Keep budgets, conversion settings and promotional conditions comparable between the control and trial.
Monitor query, asset and URL diagnostics, but avoid reactive edits that compromise the comparison.
Apply the treatment only when the result is economically meaningful and operationally acceptable, rather than relying on an isolated interface metric.

Manage automated Search through signals and exceptions

A focused weekly review should prioritize anomalies: broken conversion tracking, abrupt value changes, irrelevant query clusters, unsuitable destinations, disapproved assets and campaigns constrained by budget. These checks protect the inputs and boundaries on which automation depends.

Structural reviews can occur less frequently and cover conversion priorities, target economics, brand controls, URL rules and the experiment backlog. Bid-strategy changes need representative traffic and conversion data, so daily reactions to ordinary volatility can prevent a stable evaluation.

The advanced manager’s advantage is increasingly the ability to encode a defensible business objective, observe how automation interprets it and intervene at the level of signals, boundaries and evidence. Ahead of the scheduled September upgrade, the priority is to document the current configuration and test deliberately—not to presume that broader automation will be either an automatic gain or an automatic loss.

Google Ads’ September AI Max Switch: Audit Before Automation Expands