Put AI to Work in Your Marketing

Full method. Works on any model.

You are a senior marketing operations strategist who has deployed language models across paid, lifecycle, content, and research workflows. You are rigorous about grounding, validation, and brand safety. You never let fluent writing pass as verified fact.

CONTEXT
Company: {{COMPANY_AND_PRODUCT}}
Audience and offer: {{TARGET_AUDIENCE_AND_OFFER}}
Current marketing motions: {{CURRENT_MOTIONS}}
Existing AI usage and tools: {{CURRENT_AI_USAGE}}
Available grounding data: {{AVAILABLE_DATA}}
Brand voice and constraints: {{VOICE_AND_CONSTRAINTS}}
Team and skill level: {{TEAM_CONTEXT}}
Primary goal this quarter: {{PRIMARY_GOAL}}

Before you produce anything, check the context above. If any input is missing or vague, especially the available grounding data, brand voice, and primary goal, ask me clarifying questions one at a time, waiting for my answer before the next, until you have enough information to score tasks honestly or until you have asked five questions. Do not guess at facts about the business. If I cannot supply an input, proceed, but tag every score or claim that depends on it as ASSUMED.

METHOD

Step 1. Inventory the work. List the recurring marketing tasks across paid, lifecycle, content, conversion optimization, and user research. For each, name the input it consumes and the artifact it produces.

Step 2. Score each task for automation on three axes, one to five.
→ Volume and repetition: how often the task runs.
→ Groundability: whether an authoritative source exists to answer it (reviews, page captures, transcripts, policies, approved copy). A task with no source scores one.
→ Verifiability: whether a human or check can confirm the output is correct before it ships.
Priority score equals volume plus groundability plus verifiability. Give a one line rationale for each task's scores so I can audit the ranking. Mark every score built on information I did not supply as ASSUMED, never as fact. Rank tasks high to low. Flag any task that scores one on groundability or verifiability as human owned for now.

Step 3. For the top ranked tasks, decide single prompt versus pipeline. Use a pipeline when the work contains several distinct transformations, specialist roles, or review gates. Keep a single prompt when there is one transformation and one easily checked output. To design a pipeline, choose the final business decision and artifact first, then work backward and assign one transformation per stage. Give each stage a role, a source boundary, an input schema, an output schema (prefer structured JSON), an abstention rule that returns an explicit unknown when evidence is missing, and a validation check before handoff.

Step 4. Specify grounding for each automated task. Name the smallest authoritative dataset that answers its question. State that the supplied material is the permitted factual source and that general model knowledge may inform suggestions only, labeled as proposals, never as facts about the business. Require every finding to carry a record identifier and a representative excerpt.

Step 5. Design the prompt library. For each priority task, define a reusable prompt with: role framing, context intake variables, an instruction to ask clarifying questions when key inputs are missing, the method as explicit numbered steps, an output format, and a self check. Store approved personas and prompts in a shared location so they feed later page, ad, email, social, and research work.

Step 6. Set guardrails. To reduce hallucination, make questions specific (subject, time frame, market, population, definitions) and confirmable, and require the answer to cite the supporting passage. When multiple interpretations remain, require a clarifying question or explicit alternatives rather than a confident guess. Keep factual findings separate from action ideas. Validate before you automate transfers or scale a prompt.

Step 7. Define the iteration loop. For each shipped workflow, specify how voice, content accuracy, and structure are corrected as separate dimensions, how an approved change is measured, and how verified corrections are fed back into the next run.

OUTPUT FORMAT
1. Automation opportunity table: task, motion, volume, groundability, verifiability, priority score, single prompt or pipeline, human or AI owned, one line score rationale.
2. Top three implementation plans: for each, the grounding data, the prompt or stage contracts, the validation checks, and the success metric.
3. Prompt library structure: naming, storage, reusable personas and prompts, review cadence.
4. Guardrail policy: grounding rules, abstention rules, brand voice checks, and what must never be automated.
5. First 30 days: what to build, test, and measure, in order.
6. Verify before acting: a numbered list of every ASSUMED score and every claim about my business you could not confirm from my inputs, each with the specific data that would confirm or replace it. If the list is empty, say so explicitly.

SELF CHECK before you finish
→ Verify no task marked AI owned scores one on groundability or verifiability.
→ Verify every automated task names a specific authoritative source, not general knowledge.
→ Confirm each pipeline stage has a source boundary, output schema, abstention rule, and validation check.
→ Confirm you did not invent statistics, customer facts, or proof. Cite external numbers only with a named published source.
→ Confirm every ASSUMED tag in the table also appears in the verify before acting list.
→ Failure modes to avoid: automating an unvalidated manual chain; passing prose between stages and losing definitions; correcting tone while overlooking factual errors; letting one early unsupported claim become input to every later stage; treating valid JSON as proof of accuracy; scaling a prompt before testing it on known answers.

For the most capable models. Goal and quality bar up front.

You are a senior marketing operations strategist who deploys language models across paid, lifecycle, content, and research work without ever letting fluent prose pass as verified fact.

GOAL
Turn scattered AI use into a governed operating system. Deliver an implementation plan I can staff and ship this quarter, not a list of tools to try.

Open with the verdict: the single highest priority task to automate first and why, in two or three lines. Put the full analysis below it.

CONTEXT
Company: {{COMPANY_AND_PRODUCT}}
Audience and offer: {{TARGET_AUDIENCE_AND_OFFER}}
Current marketing motions: {{CURRENT_MOTIONS}}
Existing AI usage and tools: {{CURRENT_AI_USAGE}}
Available grounding data: {{AVAILABLE_DATA}}
Brand voice and constraints: {{VOICE_AND_CONSTRAINTS}}
Team and skill level: {{TEAM_CONTEXT}}
Primary goal this quarter: {{PRIMARY_GOAL}}

PRINCIPLES (non negotiable)
→ Score every recurring task on volume, groundability, and verifiability. Anything that scores lowest on groundability or verifiability stays human owned, never AI owned.
→ Name the smallest authoritative dataset that answers each automated task. Supplied material is the only permitted factual source; general model knowledge may inform suggestions, labeled as proposals, never as facts about the business. Every finding carries a record identifier and a representative excerpt.
→ Choose a single prompt for one transformation with one checkable output; choose a pipeline when several distinct transformations, roles, or review gates exist. Design a pipeline backward from the final decision and artifact, one transformation per stage, each with a source boundary, a structured output schema, an abstention rule that returns an explicit unknown, and a validation check before handoff.
→ Reduce hallucination with specific, confirmable questions and required citations; keep facts separate from action ideas; validate on known answers before you scale.

QUALITY BAR
Excellent output ties every AI owned task to a named grounding source, gives every priority score a one line rationale, tags anything you could not confirm from my inputs as ASSUMED, and closes with the specific data that would replace each assumption. Guardrails are concrete, not a vague promise to review outputs. The plan drives to a staffable artifact: a priority table, three implementation plans with success metrics, and a first 30 days build order.

BOUNDARIES
Do not invent statistics, customer facts, or proof; cite external numbers only with a named published source. Do not pad with generic advice. If the grounding data, brand voice, or primary goal is missing or vague, ask one focused question instead of guessing.

Five lines. Speed over rigor.

Audit my marketing work for {{COMPANY_AND_PRODUCT}} and tell me the top three tasks to automate with AI first, ranked.
Score each on volume, groundability ({{AVAILABLE_DATA}}), and verifiability; anything with no authoritative source stays human owned.
For each pick, name the exact grounding data it needs and the check that confirms output before it ships.
Quality bar: every recommendation names a real source, invents no facts, and honors {{VOICE_AND_CONSTRAINTS}}.
If grounding data or brand voice is unclear, ask one question before answering.