AI Max experiments need a result-application approval gate

Freeze the campaign question and both arms

Write the decision the experiment is meant to inform before activating it. Preserve the account and campaign identifiers, eligible campaign and exclusion rationale, control and treatment definitions, start and planned end, traffic split, budget, bidding strategy and targets, geographies, languages, audiences, search themes, negative keywords, brand inclusions and exclusions, location intent, assets, text-customization boundary, final URL and exclusion rules, landing pages, conversion actions, values, attribution settings, consent state, offline imports, audience windows, and every other campaign or measurement change during the observation period. A same-campaign design reduces some setup differences; it does not make every input stable or every observed result attributable to AI Max.

Verify the rendered setup rather than relying on the feature label. Google says the treatment enables several components by default, while brand controls added during setup apply to both arms. Record what is actually on in each arm and whether applying a shared control changed the comparison the team intended to run. Check eligibility limits, concurrent experiments, shared budgets, portfolio bidding, seasonality, promotions, inventory, pricing, site changes, tracking releases, and external media that could contaminate interpretation. If the arms, denominator, or measurement chain cannot be reconstructed, the experiment cannot authorize a campaign change.

Define favorable before the platform reports it

Predeclare the primary metric, decision denominator, minimum detectable effect, confidence or uncertainty rule, minimum runtime and conversion volume, maximum runtime, delayed-conversion window, seasonality treatment, and stopping conditions. Add commercial and customer measures appropriate to the goal: qualified lead rate, accepted opportunity, contribution margin, new-customer value, cancellation or return, sales-cycle lag, call quality, and revenue actually reconciled to the campaign. Treat modeled conversions, attributed conversions, platform value, qualified pipeline, booked revenue, and collected revenue as different evidence states. A favorable platform result can be informative without meeting the buyer's release rule.

Set non-negotiable guardrails for spend, cost per qualified outcome, brand queries, competitor or restricted terms, landing destinations, claim accuracy, customer eligibility, protected or sensitive categories, geography, accessibility, privacy and consent, complaints, sales rejection, and material anomalies. State whether a breach stops the test, blocks application, or requires specialist review regardless of the primary metric. Preserve unavailable downstream evidence as unavailable rather than zero and do not backfill a convenient metric after seeing the result. The approval packet should make it possible for an independent reviewer to reach the same apply, reject, extend, or inconclusive decision from the frozen rule and underlying records.

Keep automatic application behind named authority

Decide explicitly whether the account option to apply favorable results will remain off. For consequential campaigns, keep application manual unless a named owner has approved a narrow automatic rule with exact thresholds, guardrails, scope, expiration, notification, and rollback. Before any application, preserve the platform result, uncertainty and denominators, arm-level configuration and change history, search terms and destinations where available, conversion and value records, budget and bidding behavior, anomaly review, brand and compliance review, and the proposed base-campaign diff. Record the marketing, media, finance, measurement, product, sales, privacy, legal, and market owners whose approval is required by consequence.

Separate permission to run a bounded test from permission to change the live campaign. A media specialist may configure the experiment, an analyst may evaluate measurement quality, and a brand or legal owner may decide a specific boundary; the feature should not collapse those responsibilities into one favorable status. If automatic application is used, verify immediately that the expected components and shared controls reached the base campaign and that no unrelated setting moved with them. Missing evidence, an arm-integrity defect, an unresolved guardrail breach, or an unapproved landing or brand expansion should leave the base campaign unchanged.

Reconcile the result and monitor the released configuration

Classify the outcome as apply, reject, extend, redesign, or inconclusive and preserve the rationale. When applying, stage within a declared campaign and spend boundary, retain the pre-application snapshot, verify the exact diff, and monitor search terms, destination mix, creative variation, spend, conversion quality, delayed outcomes, brand and customer issues, and commercial results against both the experiment and prior baseline. Define a rollback threshold and owner. When rejecting or allowing settings to revert, verify that the campaign returned to the intended state rather than assuming the experiment lifecycle restored every dependency.

Reopen approval after a material budget, bid, conversion, attribution, landing page, feed, asset, brand control, location, audience, product, offer, market, privacy, or platform-behavior change. Keep platform delivery, statistically reported difference, causal interpretation, campaign configuration, qualified demand, and revenue as separate claims. Google documents a useful testing and result-application workflow, not the buyer's governance decision. The CMO remains accountable for deciding whether the evidence is sufficient, the change is authorized, the brand and customer boundaries hold, and the released configuration continues to produce the outcome the business meant to buy.

Turn this source into a reviewable decision

For AI for CMOs, use this briefing as a dated decision record rather than a substitute for the source. Preserve Google, Make AI Max work for your business with new testing and planning tools, the exact URL, the October 4, 2026 review date, the supported facts above, the editorial interpretation, the limitations, and any buyer-specific evidence. Link that record to the decisions most directly affected: Media planning and activation; Measurement and performance explanation; Brand, disclosure, and synthetic-media risk; Commerce and conversion assistance. State whether the source changes the scope, evidence requirement, control, sequence, or only the language used to describe the decision.

Before action, name the accountable owner, affected population and workflow, exact offering or configuration, source data and rights, human decision point, exception and appeal path, complete cost, expected benefit, failure and stop conditions, retained evidence, and next review date. Keep official facts, provider statements, buyer observations, representative tests, measured outcomes, editorial inferences, and unknowns visibly separate. Reopen the record when the source, offer, model, integration, data, policy, population, responsible person, or measured result changes.

Limitations and unknowns

Google is the provider and source for the August 20, 2026 AI Max testing announcement and the current Ads Help instructions. The sources support the described multi-campaign direction, same-campaign control and treatment, default treatment components, shared brand-control behavior, selected eligibility limits, and manual or favorable-result application paths. They do not independently establish a buyer's account availability, exact rendered arm configuration, traffic allocation, auction or seasonality balance, measurement validity, search-term and destination quality, favorable threshold, statistical or causal interpretation, qualified demand, margin, brand and customer effect, privacy and legal compliance, cost, or revenue. Verify current account help and interface behavior, exact experiment and campaign snapshots, change history, measurement and downstream records, guardrail review, result-application setting, post-application diff, and qualified marketing, media, measurement, brand, product, sales, finance, privacy, security, accessibility, regulatory, legal, and market-owner approval before campaign or budget reliance.