The marketing OS in the AI era
Capstone of the Marketing track. Not a new concept, it's the assembly. You put together, in a single layer, what you produced across the track: from the machine-readable catalog (N.mkt.5) to the Share of Model dashboard (N.mkt.12), with the brand audit (N.mkt.13) becoming a recurring ritual. It mirrors finance's, legal's, and sales' OS: the personal system that stays running after you close the tab and keeps improving on its own with every good delivery.
Look at your screen right now. There's a prompt saved in a notes app that always nails the post in the brand's tone. There's a company facts page you built to help AI understand who you are. There's that spreadsheet where you noted, one Sunday, how many times ChatGPT cited you versus the competitor, and never opened again. Each piece works on its own. The problem is they're scattered, depending on your memory. This lesson is the moment to put it all in one place, with a name and an order, and turn that pile of good things into a system that works for you.
Over the quarter, the marketing team produced a well-structured product catalog, a couple of campaign prompts that work, and a spreadsheet of brand mentions in AI. Each thing in a different folder, in a different person's head. When that person goes on vacation, the system goes with them. The marketing OS is the act of taking this out of one person's head and putting it on a shelf the whole team can open, with the cost logic behind it: every time someone rebuilds from scratch what already exists, the company pays an hour for not having organized it.
The legal team supporting marketing already has a verified-facts page for the brand (what can and can't be said), a claims checklist, and a log of when AI described the company incorrectly. They're scattered. The OS joins these into a layer where the facts page becomes the asset AI reads, the checklist becomes a recurring eval, and the error log becomes the audit ritual. It becomes a reputational-risk system, not three files nobody can find in time.
You spent the whole track building piece by piece: you made the catalog machine-readable, wrote the brand's facts page, built the prompt library in your tone, turned on the dashboard measuring how often AI cites you, ran the first audit of how the model describes your company. Each piece works on its own. This lesson is the assembly: you take all of it and install it as a single layer, a marketing OS, that keeps running after you close the tab and improves on its own with every good delivery.
Think of the HR team. There's an interview script that always pulls the right screening questions, a job-description template that attracts good candidates, an onboarding checklist nobody's opened in months, and a climate spreadsheet someone built on a holiday. Each thing works on its own, but lives in a different recruiter's head. When she leaves, the way of doing things goes with her. The HR OS is taking this out of the drawer and installing it on a single shelf, with a fixed home: the script becomes a named prompt, onboarding becomes an asset that runs on its own, climate becomes a recurring ritual. It stops being a pile of scattered files and becomes a system the whole team opens.
Look at your day as a product person. There's that prompt that distills discovery feedback in the PRD's tone, a prioritization template you always reuse, a product-metrics dashboard you turned on and barely open, and a roadmap document each person keeps in their own folder. Each piece is good on its own. The problem is they're scattered and depend on you remembering where you put them. The product OS is joining all of it into a single layer: the prioritization template becomes a fixed prompt, the metrics ruler becomes a recurring eval that catches regressions, and the roadmap sits in a place the whole team can find. Stop rebuilding from scratch what already exists.
Think with me about sales. There's a prospecting script that always converts better, a proposal template you tweak every time in a rush, a CRM with the pipeline, and a forecast spreadsheet someone updates from memory. Each thing works, but it's scattered and lives in a single salesperson's head. When they leave, the way of closing deals goes with them. The sales OS is putting this on a single shelf: the script becomes a named prompt, the proposal becomes a template that opens fast, the forecast becomes a ritual on top of the CRM. It stops being a scattered pile and becomes a system the whole team operates, even tired, even on vacation.
The operations team knows this well. There's a process SOP that always works, an SLA spreadsheet measuring delivery time, a quality checklist living on a lost drive, and an efficiency report someone rebuilds every Monday. Each piece works on its own, but depends on the memory of whoever built it. The operations OS is putting it all in one place, with a name and an order: the SOP becomes an asset that runs the same way every time, the quality checklist becomes a recurring eval, the SLA dashboard becomes a calendar ritual. What was tidiness in one person's head becomes a system the team can open fast.
Think of the compliance team. There's a well-built risk matrix, an internal-controls checklist validating every operation, an LGPD incident log in a folder, and an audit report template you rebuild from scratch every cycle. Each thing works on its own, but it's scattered and depends on whoever remembers where it's stored. The compliance OS is installing this into a single layer: the risk matrix becomes the asset the team reads, the controls checklist becomes a recurring eval, the audit becomes a ritual marked on the calendar. It becomes a living risk system, not three files nobody can find during an inspection.
Look at your technology setup. There's a deploy pipeline that always ships the right version, some code snippets you reuse in every project, an incident runbook nobody opens until something breaks, and an observability dashboard you turned on and forgot. Each piece works on its own, but it's scattered across repositories and in one dev's head. When they leave, the architecture knowledge goes with them. The technology OS is joining this into a single shelf: deploy becomes an automated asset, snippets become a named library, the runbook becomes a ritual on top of the dashboard. Stop rebuilding from scratch what the team already solved ten times.
Think of your UX work. There's a user-research script that always pulls the right insight, a library of prototypes and components you reuse, an accessibility checklist living lost in a file, and a journey map every designer keeps in their own Figma. Each piece works on its own, but depends on you remembering where it is. The UX OS is putting it all in one place, with a fixed home: the research script becomes a named prompt, the accessibility checklist becomes a recurring eval that catches the broken flow, the journey sits in a space the whole team can open. It stops being a scattered pile and becomes a system that improves with every usability test.
Think of the strategy team. There's a competitive-analysis template that always sorts the market out right, a scenario spreadsheet someone built for the last board meeting that nobody else can find, a deck model rebuilt from scratch every quarter, and an OKR dashboard that was set up and forgotten. Each piece works on its own, but lives scattered, in the head of whoever built it. When that person changes teams, the way of analyzing goes with them. The strategy OS is bringing this together into a single layer: the competitive template becomes a reusable asset, the scenario model becomes a named prompt, the OKR dashboard becomes a ritual on the board's calendar. It stops rebuilding from scratch, every quarter, what the team already solved before.
Let me tell you something about courses, the same way I told you in the other Business tracks. A regular course hands you a lesson: you watch, do the exercise, close the tab, and three weeks later you sort of remember the concept. Think with me: what's left in your workday? Almost nothing. A strong course is a different thing. It hands you infrastructure, something that keeps running after you close the tab, that changes how you operate on Monday morning. This entire marketing track was built to leave you with infrastructure, not memories. This lesson is where we install it for good. And notice: it doesn't bring a new concept. It's the assembly of what you've already done.
The core idea of this lesson. Your marketing OS is a living layer with four shelves: the machine-readable catalog and facts page (what AI reads to recommend you), the prompt library and content assets in your brand's tone, the brand evals (the fixed questions you run on models to measure how they describe you), and the Share of Model dashboard, with the brand audit becoming a ritual on top of it. The criterion for what becomes what is simple: whatever is repeatable and stable becomes a machine asset or recurring eval; whatever changes every time (the strategy, the claim, reading a drop) stays your judgment, with AI just helping. And here's the trick: this system improves on its own; every good campaign, every audit, leaves a deposit in the library. You're not going to leave here knowing about AI. You're going to leave with AI installed in the way you do marketing.
01The difference between a lesson and infrastructure
Let's call it what it is. What separates whoever watches an AI marketing course and keeps doing the same thing from whoever watches it and moves up a level isn't the number of prompts memorized. It's whether that turned into a system or a note.
A note is fragile. It depends on you remembering, on you finding the file, on you having the energy to rebuild the prompt in a Thursday-night rush. A system is the opposite: it's ready, it has a fixed home, it opens fast, and it works even when you're tired (or when whoever built it went on vacation). The economic question behind this is direct. How much is an hour of yours worth? Every time you rebuild from scratch something you've already done ten times, you're paying that hour for not having organized it. The OS is what stops charging you that bill.
The difference isn't magic, it's organization done on purpose. And that's exactly what we're going to do now. Fair enough?
02The four shelves of the marketing OS
The marketing OS isn't software you buy. It's a four-shelf layer you build with what you've already produced in this track. Each shelf has a clear function, and each one already has one of your pieces ready to go on it.
- Shelf 1, what the machine reads. Here lives the machine-readable catalog you organized in lesson N.mkt.5 and the brand facts page. It's what AI consults to understand who you are and recommend you. Remember what you saw about consensus: AI only recommends with confidence when it sees your brand described clearly and consistently across sources it trusts. This shelf is where you take care of that on purpose, instead of letting the model fill the gap with invention.
- Shelf 2, the prompts and assets in your tone. The collection of commands you've already tested that write in the brand's tone, plus the content templates that work. Not every prompt you've ever written. It's the subset that passed the real test, with a descriptive name, for you to reuse without rewriting (and without falling into the generic content the market already recognizes and rejects).
- Shelf 3, brand evals. The fixed set of questions that mimic your real buyer, which you run recurringly on ChatGPT, Claude, Perplexity, and Gemini to measure how they describe and cite you. It's the same eval idea you saw back in lesson 5.3 and in the Guardian track (G.6): a repeatable test that catches regressions. Here it measures brand presence, not task accuracy.
- Shelf 4, the Share of Model dashboard. The scoreboard you turned on in lesson N.mkt.12: out of your category's questions, how many times AI cites or recommends you versus the competitor. It's the shelf that gives numbers to the other three, because it's where you see whether the catalog and prompt work is actually moving the needle.
Notice this isn't theory. You've already produced a piece for each of these shelves throughout the track. The OS is the act of taking them out of the drawer and putting them on the right shelf, in the right order.
03The criterion: what becomes a machine asset, what becomes an eval, what stays in judgment
The question that trips people up most here is: what do I automate? The answer has a single criterion and it fits in one sentence, the same one you saw in the other Business tracks. The more repeatable and stable the task, the higher it climbs on the automation scale. The more it changes every time, the more it stays in your hands, with AI just helping.
- Repeatable and identical every time becomes a machine asset or eval. The brand facts page, the structured catalog, the weekly round of questions on the models. This stands on its own. You just read the result.
- Repeatable but with new content each time becomes a prompt or template. The post always has the same anatomy, but the subject changes. The prompt fixes the tone and structure and frees you to focus on the idea.
- Changes every time, stays in judgment. Positioning strategy, the claim you can or can't make, reading why Share of Model dropped this month. AI helps you think, but you're at the wheel. Trying to automate this just creates a rigid system that fails when the market changes.
This criterion spares you two costly mistakes. The first is automating what changes, and getting stuck with a machine that dumps out-of-context generic content (the AI content the public already recognizes and rejects). The second is leaving what's identical every time in your hands, and continuing to pay for your own hours out of laziness to build the eval. You want every task at the right height on the scale. Fair enough?
04The trick: the system that improves on its own
Here's the part that turns the OS from a dead file into something alive. A well-built OS doesn't sit still. It grows with every use.
Here's how it works. You run a campaign this week, and one specific post exploded in engagement and hit the brand's tone perfectly. In the old way, that work dies at publication: it worked, great, life goes on. In the OS, it doesn't die. The prompt that generated that post becomes a piece in the prompt shelf. A new fact you had to explain enters the page the machine reads. And when you run the evals and discover ChatGPT started describing your company in a skewed way, that becomes one more row in your brand audit. Every good delivery, and every scare, leaves a deposit in the system.
The compound effect of this is huge. In month one, the OS has the basics. In month six, it has your entire library of the best ways to talk to AI and to the public, distilled from dozens of real campaigns and audits. You get stronger not because AI got smarter, but because your system got more yours. The practical rule is a single one: every good campaign, and every audit, ends with a question, what's worth keeping from this? That question is what keeps the OS alive.
And notice this is the opposite of starting from scratch. Most brands start every AI-visibility fight at square one, battling to be cited again. Whoever has an OS starts from the accumulation. That's the advantage that shows up slowly and then becomes impossible to catch up to.
05The brand audit as a ritual, not a scare
Here's the moment to close the loop and give the last piece its place. The brand audit you did in lesson N.mkt.13 can't be something you do once, get a scare from, and never again. Inside the OS, it becomes a ritual: a recurring round, marked on the calendar, on top of the Share of Model dashboard.
Why a ritual and not a one-off scare? Because the way AI describes your brand changes on its own, without warning you. A new model comes out, the base it consults changes, a competitor publishes something the model starts citing, and suddenly your description changed and your Share of Model dropped (and nobody emailed you). It's the so-called dark funnel: part of the fight happens inside conversations with AI, invisible to your CRM. Without a ritual, you only discover the damage once it's already become a lost sale. With a ritual, you catch the change early, on the dashboard, and act.
And here comes a caution I can't leave out, because it's this whole module's theme, the gray edge. When you go to act on what the audit showed, the temptation is to push. Plant a hidden instruction to force the model to recommend you more. Don't do it. Remember what we saw about prompt injection back in the Guardian track (G.2): that's manipulation, it violates anti-spam policy, and it has a perverse, research-documented effect (the "injection paradox"): in safety-trained models, the classifier detects the attempt and silently downranks your entire brand. You end up worse than if you hadn't touched anything. The OS's path is the light side: clean catalog, verified facts, consistent presence in trusted sources. The audit shows the hole; you fill it with truth, not with a trick.
If you want to see where the marketing OS fits into the bigger picture, it's your personal instance of the same infrastructure finance, legal, and sales built in their own tracks. Marketing was just the domain where you built this one. The method is the same for any area: gather the pieces, define what climbs the automation ladder, and keep it alive with use.
To take with you: your marketing OS has four shelves (what the machine reads, the prompts in your tone, the brand evals, and the Share of Model dashboard), and the brand audit becomes a ritual on top of it, not a one-off scare. The criterion is repeatability: the stable climbs to a machine asset or eval, what changes stays in your judgment. The system improves on its own because every good delivery leaves a deposit. And you fix what the audit shows with truth (catalog, facts, consistency), never with hidden injection, which downranks the whole brand. Fair enough? Next up.
Do it now
Open a blank document and title it: Marketing OS, your real task. Create the four shelves as sections and pull in what you've already produced in this track:
- What the machine reads. Paste (or link) the machine-readable catalog from lesson N.mkt.5 and your brand facts page. In one line, write: could AI describe me correctly with just this? If the answer is "it would have to guess", mark it as pending.
- Prompts and assets in your tone. List three to five prompts you tested in this track that write in the brand's tone. Give each a descriptive name (e.g., "authority post in the brand's tone", "objection response on LinkedIn") and paste the prompt.
- Brand evals. Write five to ten questions your real buyer would ask AI in your category (e.g., "what's the best tool for X for a small company?"). This is your evals library, which you'll run recurringly on ChatGPT, Claude, Perplexity, and Gemini.
- Share of Model dashboard. From lesson N.mkt.12's dashboard, note down today's number: out of shelf 3's questions, in how many does AI cite or recommend you, versus the competitor. This is your starting mark.
At the end, mark the brand audit (N.mkt.13) on the calendar as a recurring, monthly ritual, on top of this dashboard. And write, in one sentence, the light-side rule: when the audit shows a hole, I fill it with truth (catalog, facts, consistency), never with hidden injection. Starting today, every good campaign ends with the question: what's worth keeping from this?
Why Share of Model is becoming the KPI that's replacing Share of Voice
The old brand metric was "share of voice": what slice of your category's mentions, in media and search, was yours. In the AI era, a sibling emerged, Share of Model (SoM): the slice of AI's answers, for your category's questions, in which your brand is cited or recommended as the main solution. Market analysts have been pointing to SoM as the 2026 KPI precisely because ranking well on Google stopped guaranteeing you show up in AI's answer, and much of the journey now happens inside the conversation with the model, invisible to traditional Analytics (the dark funnel). There are fine variations on the concept (AI Share of Voice, which measures only your slice of mentions; and Share of Answer, which measures how much of the answer itself is about you), but the central idea is the same: stop measuring only clicks and start measuring presence in the answer. You measure this by recurringly running your evals library across several engines, calibrating lab data (controlled prompts you fire) with field data (tools observing real clickstream, like Semrush, Profound, Datos). You don't need to become a metrics engineer. You need to recognize that "being in the answer" became the scoreboard, and have a ritual that reads it every week.
Practice
1. What's the criterion for deciding what becomes a machine asset/eval, what becomes a prompt, and what stays in your judgment, in the marketing OS?
2. Why should the brand audit (N.mkt.13) become a recurring ritual inside the OS, instead of a one-time check?
3. The audit showed ChatGPT has been describing your brand weakly and citing the competitor. Which action is aligned with the marketing OS?
Fair enough? Let's close the point together. This whole track was built to leave you with infrastructure, not memories. Your marketing OS joins into a single layer what you built piece by piece: the catalog and facts the machine reads, the prompts in your tone, the brand evals, and the Share of Model dashboard, with the audit becoming a ritual on top of all of it. The criterion is repeatability: the stable climbs to automation, what changes stays in your judgment. The system improves on its own because every good campaign leaves a deposit, and you fix what the audit points out with truth (catalog, facts, consistency), never with hidden injection that downranks the whole brand. You didn't finish an AI-marketing course, you installed a marketing that operates in the AI era, and nobody can take that away from you.
For the board
On the criterionthe facts page and the eval runs move to automation. The post becomes a prompt. Reading a drop stays in your judgement.
On the rituala new model, a shifting base, a competitor that starts getting cited: all of it changes your description without telling you.
On the fixgive the model a clear and consistent identity so it stops filling the gap with invention.
Thanks for the feedback. It helps sharpen the next lesson.