Master Prompts for Marketing: Campaign Briefs, Copy, and Performance Reports with Structured AI Output

Marketing AI becomes risky when fast copy separates a claim from its evidence, an audience from its permission, or a performance statement from its measurement method.

A marketing master prompt should package approved strategy and facts into structured drafts. Marketers, product owners, analysts, legal reviewers, and channel systems retain control of what is claimed, targeted, measured, and published.

Build the Campaign Brief First

{
  "campaign_id": "approved-id",
  "objective": "approved-objective",
  "offer_and_product_scope": [],
  "audience_definition_id": "approved-segment",
  "approved_claim_ids": [],
  "required_disclosures": [],
  "channel_and_market": {},
  "measurement_plan_id": "approved-plan"
}

The model should not invent the objective, audience, offer, claim, proof, or success metric while writing the campaign.

Govern Claims as Data

Each objective claim needs approved wording, evidence, product and variant scope, market, channel limits, required qualification, owner, and review date.

The FTC says advertising claims must be truthful, non-deceptive, and evidence-based, and advertisers need a reasonable basis for objective claims before dissemination. See the official FTC advertising and marketing guidance.

The model may use approved claims but should not strengthen “may help” into “will,” broaden test results, create a comparison, or imply scientific support that is not in the record.

Generate Channel-Specific Drafts

Return separate subject, preview, headline, body, call to action, social copy, ad fields, landing-page sections, and metadata under the channel's versioned constraints.

{
  "draft_assets": [],
  "claim_references": [],
  "disclosure_placements": [],
  "unsupported_statements": [],
  "channel_validation": [],
  "marketing_review_required": true,
  "approved_for_publish": false
}

Code checks lengths, required fields, prohibited terms, destination, links, offer dates, and disclosure presence. Human review considers the overall impression, not just individual phrases.

Audience and Personalization

Use approved segments and permitted attributes from governed systems. Do not infer sensitive characteristics or create lookalike logic inside the prompt.

Consent, suppression, frequency, age, geography, customer status, and channel eligibility should be enforced before generation and again before sending. The model cannot create permission.

Personalization should use facts that improve relevance, not hidden judgments about vulnerability, health, finances, identity, or willingness to pay.

Endorsements and Reviews

Keep the actual endorsement or review, permission, material-connection disclosure, usage scope, date, and any substantiation for underlying product claims. Do not fabricate testimonials, create a customer experience, or remove a qualification.

The FTC explains that endorsements must be truthful and not misleading and material connections should be disclosed. See the official FTC endorsement guidance.

Performance Reports

Analytics systems calculate impressions, reach, spend, conversions, revenue attribution, experiment results, and confidence under approved definitions. The model may explain verified metrics and supplied context.

Do not let it choose a favorable attribution model, claim causation from correlation, hide missing tracking, or compare periods with changed definitions.

Every metric needs source, window, timezone, filters, calculation version, and known limitations. Separate observed result, experiment conclusion, analyst interpretation, and proposed next action.

Experiments Need Predefined Methods

Record hypothesis, variants, audience, allocation, start and stop rules, primary metric, guardrail metrics, exclusions, and analysis method before interpreting results.

The model can draft a summary from verified experiment output. It should not move the stopping point, select a different metric after seeing results, or declare causation when the design does not support it.

Correct at the Same Scale

When a claim, offer, link, audience, disclosure, or metric is wrong, identify every affected asset and channel. Assign a correction owner, replacement version, publication status, and customer-notice decision.

Do not merely fix the prompt. Content already published, cached, syndicated, emailed, or handed to sales needs its own correction record and confirmation.

Control Publication and Spend

The model should not launch campaigns, change bids or budgets, publish content, alter audiences, send messages, approve offers, or respond publicly. It returns drafts and proposals.

Application code verifies current assets, approvals, claim status, consent, channel configuration, spend authority, and idempotency before any action.

Test Marketing Harm

Test unsupported claims, expired evidence, wrong variant, missing disclosure, fabricated testimonial, opted-out recipient, sensitive proxy, changed offer, broken link, unavailable product, ambiguous experiment, prompt injection in customer content, and request to bypass review.

Read Master Prompts for Retail for product-source controls and Master Prompts for Tech Teams for versioned documentation pipelines.

Marketers Own the Message

Marketing leaders own strategy, audience, positioning, channel, and publication. Product owners own product facts. Analysts own measurement methods and interpretation. Legal, privacy, and compliance teams own applicable review. Developers own identity, source provenance, schemas, validation, permissions, and actions.

The master prompt creates consistent campaign artifacts. It does not create truth, permission, or strategy.

Browse marketing workflow contracts in the CyWire marketplace.

This article is technical information, not advertising, privacy, consumer-protection, financial, regulatory, or legal advice.