Data contract for AI Max campaign steering: fields, freshness, ownership and QA

Short answer: Treat AI Max campaign steering as governed configuration data. Store messaging, matching and audience guidelines as versioned fields with owners, effective dates, campaign scope, review status and rollback state. Keep previews and observed delivery separate from intended instructions so the team can distinguish what it asked the system to do from what actually served.

Why steering instructions need a data model

Google's AI Max update describes AI Brief as a way to provide business context through messaging, matching and audience guidelines, with previews that let advertisers iterate before committing.

Those instructions are not just prose. They affect campaign behavior and therefore need the same discipline as other material configuration.

A data contract makes them inspectable across media, brand, legal and measurement teams.

Core identity fields

Every steering record should identify:

• account;
• campaign;
• campaign type;
• market;
• language;
• owner;
• effective date;
• status;
• version;
• previous version;
• rollback version.

Without identity and versioning, a later performance shift can be impossible to reconcile with the instructions that were active at the time.

Messaging-guideline fields

Store messaging instructions in structured categories where possible.

Useful fields include:

• message_priority;
• required_claims;
• prohibited_claims;
• mandatory_disclaimer;
• price_language_policy;
• brand_terms;
• tone_notes;
• evidence_required;
• review_owner.

Avoid one large free-text block when several independent constraints can be represented separately.

Matching-guideline fields

Matching guidance should make the business boundary explicit.

Possible fields:

• priority_intent;
• excluded_intent;
• brand_scope;
• geographic_scope;
• regulated_topic_exclusion;
• overlap_campaigns;
• negative_keyword_dependency;
• search_term_review_owner.

The goal is not to recreate a keyword list. It is to preserve the commercial boundary the automation should respect.

Audience-guideline fields

Audience guidance can include:

• audience description;
• business reason;
• allowed personalization theme;
• prohibited personalization claim;
• market eligibility;
• privacy constraint;
• outcome metric;
• review owner.

Keep audience descriptions grounded in permitted targeting and business logic rather than inferred sensitive characteristics.

Preview evidence

Google says AI Brief can show sample assets and searches before commitment.

Store preview evidence separately from the instructions:

• preview timestamp;
• sample asset text;
• sample search context;
• reviewer;
• accepted/rejected;
• rejection reason;
• revision requested.

A preview is an example, not proof of all future delivery.

Observed-delivery fields

After launch, collect the actual evidence available from campaign reporting.

Useful fields include:

• observed search-term theme;
• served asset;
• landing page;
• policy issue;
• brand-review issue;
• conversion outcome;
• downstream quality where available;
• observation period.

This layer lets the team compare intended steering with actual behavior.

Freshness rules

Campaign steering becomes stale when the underlying business changes.

Trigger review when:

• price policy changes;
• product availability changes;
• market changes;
• legal language changes;
• brand positioning changes;
• campaign objective changes;
• landing-page architecture changes;
• product rollout changes platform capability.

Do not leave old guidelines active just because performance reporting still looks stable.

Ownership model

Assign explicit ownership:

• business owner: product and offer truth;
• brand owner: messaging constraints;
• media owner: matching and campaign scope;
• legal/compliance: mandatory/prohibited language;
• measurement owner: outcome interpretation;
• operator: configuration and reconciliation.

A single person can hold multiple roles, but the fields should remain conceptually separate.

QA checks

A pre-launch QA routine should detect:

• contradictory instructions;
• expired offer language;
• audience that cannot use the product;
• prohibited price claims;
• missing disclaimer;
• campaign overlap;
• undefined rollback version;
• previews that conflict with product truth;
• changed landing pages not reflected in the steering record.

Safe lifecycle states

Use states such as:

• DRAFT;
• REVIEW_REQUIRED;
• PREVIEW_ACCEPTED;
• ACTIVE;
• OBSERVATION_REQUIRED;
• REVISION_REQUIRED;
• ROLLED_BACK;
• RETIRED.

These states help the team know whether a steering instruction is proposed, active or historical.

Reconciliation after a steering change

After any material update, compare the stored instruction set with the configuration visible in the account and with the first reporting window. Record whether the change was fully applied, partially reflected or still unknown. If preview behavior, served assets or search-term patterns conflict with the approved steering record, open a reconciliation item instead of silently editing the historical version. This preserves a trustworthy timeline for later performance analysis and rollback.

The data-contract rule

AI Max steering should be managed as versioned campaign configuration with evidence around it.

Store what the team intended, what the platform previewed, what actually served and who approved each stage. That makes automation steerable and auditable rather than conversational but opaque.

Sources reviewed

• https://blog.google/products/ads-commerce/ai-max-new-features/