AI marketing engineer · build, not advice
AI marketing engineer: the machinery, not the deck
Most “AI marketing” is a subscription plus an opinion. An AI marketing engineer does something narrower and checkable: takes the path that is supposed to bring customers and makes it run — agent skills that produce and check the work, pages built so answer engines can quote them, sequences and routing that fire on their own, and an event that records what happened. Packages are published before the call: Sprint $900, Engine $1,900/month, Full build $2,900.
Four things that have to exist before “AI marketing” means anything
A page a model can quote
A page that renders without JavaScript, answers one question in the first 60 words, states prices as numbers and carries Service/Offer or FAQPage markup. Writing the page for a human reader and a model reader is the same job. Free check: 23-point readiness checklist.
Skills, not prompts
Good output comes from a reusable operating file the agent loads every time — what to do, what never to do, which numbers are floors and not opinions. Ours is public: marketing-mindset, also on npm and as an MCP server.
A kill rule written first
How many data points, at what result you stop, what happens to the winner. The sample-size floor for a two-variant email test is roughly 1,500–2,000 sends per arm; the kill-rule calculator returns the Wilson interval and the rule-of-three floor when conversions are zero.
What “AI marketing engineer” has to beat
| AI marketing engineer | The usual “we use AI” retainer | What you can open afterwards |
|---|---|---|
| The unit of output | An AI tool subscription | An asset in your accounts, with a link |
| Where the AI sits | In the chat window | In the pipeline that writes, checks and ships |
| What a month leaves behind | Screenshots of prompts | Pages, events, sequences, a changelog |
| Can a model quote you? | Not addressed | Schema, llms.txt, plain-HTML answers |
| When it stops working | You cancel the seats | The machinery keeps running; you own it |
The test is one question: name something I will be able to open when the month is over. A subscription answers with a tool. An engineer answers with a URL, an event and an account.
Ships you can open right now (the method applied to itself)
- This page — Service/Offer + FAQPage schema, GA4 event on the Telegram click.
- axelfreeman/marketing-engineer-playbook — the method in five chapters, MIT.
- npm: marketing-engineer-playbook
— the same method as a CLI (
npx marketing-engineer-playbook read 5). - Test planner — how many sends an honest test needs.
- Marketing engineer job description generator — free tool: the JD with artifacts due by day 90, the sample-size floors and the screening questions.
- Proof page — every artifact of this engine with a live link, dated.
- Marketing automation engineer — capture, routing, suppression and failure alerts.
- Marketing engineer services — the seven deliverables, each with something you can open before paying.
- How to become a marketing engineer — five steps, each with something you can open: one channel, a measurement stack, three dated public artifacts.
- Marketing engineer vs freelancer — what you are actually buying: who owns the number, how the work is checked, and what you keep at handover.
- Marketing engineer vs growth marketer — what each role owns by default, the first readable result, and four signs that decide the hire.
- Marketing engineer for agencies — an agency cannot pool its clients' traffic, so every client needs its own floor and its own read-out date: 13,914 per arm (27,828 total), 14 months for a client at 2,000 contacts a month, and 4,638 contacts a month as the smallest client readable inside 6 months. li>Marketing engineer for marketplaces — two funnels that must be readable at the same time, liquidity as the readable unit, why each side needs its own floor (13,914 per arm; 27,828 in total; 70 months for a 400-seller-a-month supply side) and the four decisions taken before the first contact.
- Marketing engineer interview questions — the six questions that separate an engineer from a marketer with tools, what each answer really measures, the volume a verdict needs (13,914 contacts per arm; 27,828 in total), and the four things the interview cannot decide. li>Marketing engineer for devtools — why the readable unit is the first successful call rather than the signup, the volume a verdict needs, and the stop rule that has to exist before the spend.
- Marketing engineer for fintech — why the readable unit sits after identity and funding, the volume a verdict needs, and the stop rule that has to exist before the spend. li>Marketing engineer retainer — the invoice line by line, the four decisions a retainer cannot make for you, and how many honest tests a month can actually hold at your volume.
- Marketing engineer job description — the four decisions and six lines to write instead of a tool list, the artifact due each week, the volume a strict test needs, and a free generator. li>Marketing engineer for B2B — the lead, source and reply definitions that decide whether a channel can be judged, the five things that exist before the first send, and the floor an honest test needs. li>Marketing engineer for SaaS — the five numbers most plans are missing, what gets instrumented first, and what ships in a $900 Sprint.
- Marketing engineer for startups — the three routes a seed-stage team compares, what the first 30 days produce, and the floors the numbers have to clear.
- Marketing engineer for ecommerce — why last-click revenue is not a verdict, the spend gate, and the traffic floors a store test has to clear.
- AEO service — being quotable by answer engines, done as work rather than as an audit.
How the engagement works
One path, live in a week
Pick the broken or unmeasured path: the page, the sequence or the reporting. It gets built, instrumented and running, and you keep it.
The continuing block
Build, measure, kill the losers, scale the survivor. Fourteen days notice to cancel — a retainer you cannot leave is a lease, not a service.
Handed over running
Offer, pages, outbound, automation, measurement and documentation, so the system survives without the operator.
Questions people actually ask
What is an AI marketing engineer?
The person who makes marketing work without a human in the loop: agent skills that write and check the copy, pages built so answer engines can quote them, sequences and routing that run themselves, and tracking that reports what happened. The word “AI” is not the job title of a prompt writer — it is the unit of output: a working asset with an event attached, not a plan about AI.
How is an AI marketing engineer different from an AI marketer or an AI consultant?
An AI marketer talks about tools. A consultant writes a strategy and leaves. An engineer ships inside your own accounts: the page, the schema, the llms.txt, the sending domain, the event, the kill rule. When the engagement ends, the machinery is still in your accounts and still running.
What does the work actually produce?
Openable artifacts. On this site the same method produced an offer page with Service/Offer and FAQPage markup, a 56-URL sitemap with IndexNow pings, an llms.txt written for model crawlers, three free browser tools, four published npm packages, a public playbook repository and two dozen pages written under real queries. Every one of them is listed with a live link on the proof page.
How do I know the AI part is not decoration?
Check whether the artifacts can be opened and run. The npm packages are on the registry, the tools run in the browser with no account, the playbook chapters are readable on GitHub, the sitemap and the snapshot of each page are public. If a vendor cannot show a link for the thing they claim, the claim is a deck.
How is it priced?
Three published packages, no hourly rate: Sprint $900 one-time (one path built, instrumented and live), Engine $1,900 per month (build, measure, kill or scale; 14 days notice), Full build $2,900 one-time (the whole system handed over running). The comparison with a freelancer, an agency and a full-time hire is on the cost page.
What is out of scope?
Ad spend management, content volume promises, hiring, and paid licences. Ad spend and subscriptions stay in your accounts, with no markup. The scope in writing: scope of work.