M11.L1 · AI, Ethics & Law
AI as your marketing operating system
What you'll be able to doEvaluate which of your recurring marketing tasks AI should accelerate, which it should draft for review, and which it must not own — producing a task-by-task capability map.
Learn
What changed, and what didn't
Module 5 made an argument this lesson builds on and will not contradict: AI made average content free, which made it worthless. Nothing in Module 11 walks that back. If you came here hoping the machine would write your marketing for you, Lesson 5.1 already told you what that output is worth, and the market agrees.
But that argument was about output. This module is about operations — the hours between the ideas. Most of a working marketer's week is not the scarce, only-you-could-make-this work Module 5 demanded. It is summarising, drafting, reformatting, categorising, chasing. That layer is exactly what AI is good at, and refusing its help there is not integrity — it is spending your scarcest resource, hours, on work that doesn't need your judgement.
So the frame for this module: AI is not your marketer. It is your marketing's operating system — the layer that runs the repetitive work while you keep the three things it cannot hold: judgement, taste, and responsibility for what's true.
The four strong suits
Where the machine genuinely earns its place:
Research synthesis. Give it forty customer reviews, six interview transcripts, or a competitor's pricing page, and ask for recurring themes. It is summarising text you supplied — its strongest mode, because there is nothing to invent. Your Module 2 clustering work, done in a tenth of the time. You still judge the clusters.
First drafts. A structured brief in, a competent draft out. Not a publishable draft — a draft that spares you the blank page. The distance from blank page to bad draft is where most writing time goes.
Variants at volume. Ten subject lines, fifteen ad hooks, five headline framings of the same offer. Generating options is cheap for the machine and expensive for you, and Module 9 taught you the market picks the winner anyway.
Structured analysis. Categorising a support-inbox export, tagging survey answers, turning a messy spreadsheet into a tidy one. Boring, error-prone by hand, mechanical by machine.
Notice what all four share: you provide the raw material, and a human judges the result. That is not a coincidence. It is the design rule.
The four failure modes
Facts. The machine produces fluent, confident, wrong statements — invented statistics, invented features, invented people — and nothing in the prose signals which sentences are the invented ones. This is the failure mode with teeth, and Lesson 11.2 builds a whole discipline around it.
Taste. Left to itself, it produces the median of everything it has read: competent, generic, could-be-anyone. Module 5 already showed you what that's worth.
Strategy. It optimises the goal you give it. It will never tell you the goal is wrong, the channel is wrong, or that the honest answer is "don't run this campaign". Strategy is choosing; the machine has nothing at stake.
Brand voice. Without strong examples it drifts to the internet's average register — enthusiastic, vague, faintly American. Your Module 3 voice work exists precisely because that register sells nothing.
Accelerate, Draft, or Own
The practical tool of this lesson is a three-way classification for every recurring task:
- Accelerate — AI does it, you spot-check. Right for synthesis and mechanical work where errors are cheap to catch.
- Draft — AI produces the first version, you rewrite. Right for anything published in your voice.
- Own — human only. Right for strategy, judgement calls, and any claim of fact that faces a customer.
Two forces decide the classification. First, which failure mode the task exposes. Second, the verification burden: every factual claim the machine makes costs checking time. For a task that is mostly facts — a case study, a price list, a legal page — checking the machine's version can take longer than writing your own. For factual work, time saved by AI can be genuinely negative. That is not a reason to avoid AI; it is a reason to classify honestly.
Context is the multiplier
The single biggest lever on output quality is not the prompt's cleverness. It is what you feed in. A request for "a LinkedIn post about client onboarding" produces sludge. The same request with your ICP from Module 2, your positioning statement from Module 3, three real customer quotes and two of your best past posts produces something recognisably yours, one edit away from usable.
You have spent ten modules building exactly those artefacts. They were always your strategy. Now they are also your fuel — which is one more reason a student who skipped the research modules gets generic output and concludes "AI doesn't work for my business". It works fine. It just had nothing of the business to work with.
Lena maps a month
Lena has about 6 hours a week for marketing — roughly 24 hours a month. Her audit of where they go: about 10 hours drafting her two-a-week LinkedIn posts and the monthly newsletter; about 4 hours summarising demo-call transcripts from the larger firms; about 3 hours inching the Van Dijk case study forward; the rest scattered across admin. Her map:
Transcript synthesis → Accelerate. The machine summarises text she supplies; she checks each summary against the transcript in 15 minutes. Saves roughly 3 hours a month. (One caveat she writes down for Lesson 11.3: transcripts contain customers' personal data, and where they go matters legally.)
LinkedIn posts and newsletter → Draft. Variants generated from her positioning doc and best past posts; she rewrites in her own voice. The machine never publishes.
The case study → Own the claims. Structure and tightening, fine. But every number and every customer-facing statement is hers. She tested why: asked for a draft with only the consented Van Dijk quote as input, the machine returned a version that added "reduced onboarding time by 73%" and a second firm, "Van Dael Accountancy". Neither exists. A hallucinated figure in a published case study is not an embarrassment — it is the end of a customer relationship.
The Ostara contrast makes the same boundary from the other side. Mateus lets AI draft 40 ad-copy variants from the message-house pillars — cheap to verify, and the market tests them anyway. But glaze and food-safety claims are Own, always: Ostara's glazes are lead- and cadmium-free, tested for food contact to EN 1388, and every published word about that must trace to the certificates on file. Regulated fact is never a drafting exercise.
The centaur principle
One sentence to keep: the human sets direction and judges; the machine generates and processes volume — never the reverse. The losing patterns are both inversions. Publishing first drafts unread is letting the machine judge. Refusing all of it is making yourself the volume processor. Neither survives contact with a competitor who got the division of labour right.
Now map your own tasks.
Do
Exercise 11.1.1 — Your AI capability map
Audit your five most time-consuming recurring marketing tasks and classify each: Accelerate, Draft, or Own — with your reason, and the failure mode you're guarding against.
Write these down — in your plan document, or on the worksheet at the end of this lesson.
| What to write | Guidance |
|---|---|
| Your five tasks | Exactly five recurring tasks, each with: the task in a phrase · hours per month · its classification (Accelerate · Draft · Own) · your reason in 20–60 words · the failure mode you're guarding against (facts · taste · strategy · voice · none) |
| The two-way check | At least one task must be Draft and one Own. "AI does everything" and "AI does nothing" are both wrong maps — the first hands judgement to the machine, the second makes you the volume processor |
| Hours you expect back | Add them up, then subtract checking time. If the estimate exceeds 60% of your total hours, verification time is real time — recount |
| Your first pilot | One of the five: the task you will build a written SOP for in Lesson 11.2. Don't pick an Own task |
Once classified, sort your five tasks into three columns with monthly hours per column — that is the shape of your own operating system.
Where this goes: Operations §1 — AI Usage — of your Marketing Plan. Your pilot choice is the starting point of the Lesson 11.2 SOP.
Check
Rubric
Mark your own work against these criteria.
| Criterion | 8–10 | 5–7 | 1–4 |
|---|---|---|---|
| Task realism | Five genuinely recurring tasks with plausible hours | Tasks real but hours guessed silently | Vague activities ("do marketing") or invented hours |
| Classification defensibility | Each classification argued from the task's actual failure mode | Broadly right, reasons thin | Facts or strategy handed to Accelerate, or everything marked Own out of caution |
| Failure-mode awareness | Named failure modes match the tasks (facts for claims, voice for published copy) | Modes named but generic | None everywhere, or modes contradict the classification |
| Pilot choice | Pilot is a recurring Accelerate or Draft task with real hours behind it | Defensible but low-value pilot | Pilot is an Own task, or a one-off |
Pass: 5+ on every criterion. Distinction: 8+ on all four.
Quiz — 4 questions
1. Lena asks AI to draft a case study and it produces "Storkflow cut Van Dael Accountancy's onboarding time by 73%" — a firm and a figure that don't exist. Which failure mode is this?
- a) Taste — the output is too generic
- b) Facts — fluent hallucination; customer-facing numbers must be human-owned ✔
- c) Strategy — the wrong goal was optimised
- d) Voice — it doesn't sound like Storkflow
Why: the sentence is confident, specific and false, and nothing in its fluency signals the fabrication. That is exactly why claims of fact sit in Own, and why Lesson 11.2 bans any figure you cannot trace to your own inputs.
2. Which task is the best fit for AI acceleration with a light human check?
- a) Deciding Ostara's positioning for next year
- b) Summarising 40 customer reviews into recurring themes ✔
- c) Approving a food-safety claim about a ceramic glaze
- d) Setting the marketing budget
Why: (b) is synthesis of text you supplied — the machine's strongest mode, with cheap verification. The others are strategy, regulated fact, and judgement: two Owns and a decision no machine has anything at stake in.
3. A marketer reports that using AI to write a fact-heavy comparison page "saved no time — checking everything took longer than writing it". What does this show?
- a) They used the wrong AI tool
- b) Comparison pages shouldn't be written at all
- c) The verification burden — for factual work, checking AI's claims can cost more than producing your own, which is why such tasks classify as Draft or Own ✔
- d) AI output never needs checking if the prompt is good
Why: every factual claim the machine makes is a claim you must verify before publishing. When a task is mostly facts, that overhead can exceed the saving — an argument for honest classification, not for abandoning the tool.
4. How does this lesson relate to Module 5's argument that AI-generated volume is worthless?
- a) It replaces it — AI has improved since Module 5
- b) It contradicts it — volume is fine if the prompts are good
- c) It extends it — average output is still worthless, so AI runs the operational layer while humans supply the scarce material and the judgement ✔
- d) They are unrelated
Why: Module 5 was about what to publish; this module is about how the work gets done. The division of labour is the same in both: you bring the ore, the machine refines — and it still refines nothing into nothing.
Advance
You now have a map most working marketers never draw: which of your hours the machine should take, and which it must never touch. The estimate beside it — hours back per month — is the budget for everything else this Program asks of you.
Next: M11.L2 — Repeatable AI workflows. A good prompt used once is luck. Next lesson turns your pilot task into a written procedure anyone could run next week — including you, in six months, on a bad day.
Mark your own work
| Good | Not yet | |
|---|---|---|
| Tasks are real | Five recurring tasks with hours attached | "Content" as a task, hours invented |
| Boundaries drawn both ways | At least one Draft and one Own, argued | Everything to AI, or nothing |
| Failure modes named | Each guard matches its task | None on a task full of factual claims |
| Verification counted | Hours-saved estimate survives the recount | Checking time assumed to be zero |
Worksheet
THE SCHOOL OF NET MARKETING
Lesson 11.1 — AI as your marketing operating system
THE RULE
Human sets direction and judges.
Machine generates and processes volume.
Never the reverse.
MY FIVE BIGGEST RECURRING TASKS
Task h/month A / D / O Guarding against
1. ____________________ ______ _________ ☐facts ☐taste ☐strategy ☐voice
2. ____________________ ______ _________ ☐facts ☐taste ☐strategy ☐voice
3. ____________________ ______ _________ ☐facts ☐taste ☐strategy ☐voice
4. ____________________ ______ _________ ☐facts ☐taste ☐strategy ☐voice
5. ____________________ ______ _________ ☐facts ☐taste ☐strategy ☐voice
Accelerate = AI does it, I spot-check
Draft = AI first version, I rewrite
Own = human only — strategy, judgement,
any factual claim a customer sees
HOURS I EXPECT BACK PER MONTH: _______
(Subtract checking time. If this is more than
60% of the total, recount.)
MY FIRST PILOT (one Accelerate or Draft task):
____________________________________________
SELF-CHECK
☐ At least one task in Draft and one in Own
☐ Every factual/customer-facing claim sits in Own
☐ My failure modes match my tasks
☐ My pilot recurs — an SOP for it will pay weekly
Next: Lesson 11.2 — Repeatable AI workflows.
theschoolofnetmarketing.com/learn/repeatable-ai-workflows