Prepare a ChatGPT Ads test
Start with: an approved offer, audience, budget, landing page, and measurement plan.
Stop when: the offer, tracking, platform access, or approval owner is unresolved.
DIY AI marketing guides for operators
Use this directory to choose a bounded AI marketing task, identify the evidence it needs, and decide what still requires human judgment.
What this directory owns: AI-assisted marketing operations, including ad preparation, landing-page checks, customer-list handling, citation-answer drafting, and inquiry review. For AI search visibility, technical SEO, schema, and answer-engine measurement, use the separate DIY AI SEO directory.
Choose the job you need to do
Handle customer-list work
Start with: documented consent, list origin, permitted use, retention rules, and a named data owner.
Stop when: consent, deletion, access, or provider terms cannot be verified.
Make an answer citable
Start with: a buyer question, a direct answer, a primary source, an accountable reviewer, and a date.
Stop when: the claim has no source or the page would duplicate the AI SEO directory.
Review inquiry quality
Start with: the original inquiry, its source, qualification criteria, response history, and privacy-safe notes.
Stop when: the model is being asked to make an eligibility, pricing, legal, or relationship decision without human approval.
What AI can assist with, and what stays human
Use AI to assist
Summarize supplied evidence, draft variants, classify clearly defined inputs, and surface missing information. Keep the source material and review trail.
Do not delegate
Do not let a model invent proof, approve claims, decide consent, set policy, or make a consequential customer decision. A fluent answer is not evidence.
Require approval
Name the person who checks factual claims, privacy, brand fit, platform rules, measurement, and the final go-live decision.
Operating references: NIST AI Risk Management Framework, FTC guidance on AI claims, and Google Search guidance for generative AI content.
Copy this AI marketing task contract before opening a tool
Outcome and evidence
Outcome: name one business or operating result. Evidence supplied: list the approved source files, customer language, facts, and current platform documentation.
Allowed work and prohibited decisions
AI may: summarize, classify, compare, or draft the named output. AI may not: invent proof, approve a claim, decide consent, set price, or make a consequential customer decision.
Reviewer and approval record
Name the factual reviewer, privacy or policy reviewer when needed, final approver, decision date, approved version, and any required correction.
Measure and stop condition
Define accepted output, time saved, rework or error count, qualified action, and downstream result separately. Stop when evidence, authority, access, consent, or rollback is unresolved.
Match review effort to the decision risk
Low-risk assistance
Summaries, outlines, formatting, and variants made from supplied material. A competent human checks accuracy and brand fit before use.
Medium-risk publication or spend
Public claims, ads, landing pages, audience rules, and campaign changes require source review, platform checks, a named approver, and a reversible launch plan.
High-risk decisions
Consent, legal status, eligibility, pricing exceptions, safety, employment, credit, and relationship decisions stay with an authorized human and the applicable policy owner.
DIY or ask for help?
Continue DIY
The task is bounded, the evidence is available, the reviewer understands the subject, errors are reversible, and one owner can approve the result.
Pause and get help
The work crosses several systems, changes paid spend or customer treatment, uses sensitive data, lacks a competent reviewer, or cannot be rolled back safely.
Worked example: draft a buyer-question answer without inventing proof
An operator supplies five actual customer questions, the current return policy, product specifications, and the responsible page URL. AI groups duplicate questions and drafts one answer per question. The factual reviewer checks every specification and policy statement against the supplied sources. The page owner approves the final language and records the version. If the return policy is missing or contradictory, drafting stops. The outcome record separates time saved, corrections required, published answers, and later qualified actions; it does not call faster drafting a sales result.
Full page directory
Show all pages
Use the guides when the task, evidence, risk, and approval owner are clear. If several systems or decision owners are involved, See marketing services
AI Citation covers the underlying rule this depends on.