Using AI for Marketing: What Two and a Half Years of Daily Use Taught Me

I do my job on two computers.

On the first one, I’m the head of marketing at SSI SCHÄFER Switzerland, a multinational in intralogistics. That machine lives inside strict corporate AI policies. Frontier models are off limits for company data, because practically no AI company today can guarantee, with 100 percent certainty, what happens to the data you feed them. What runs instead is disciplined: marketing analysis on cleaned, anonymized data where policy allows, and predictive capabilities through established vendor tools, which I’ve assessed honestly elsewhere.

On the second one, I’m a solo marketing consultant. On that machine, practically everything runs through AI. Research, writing, analysis, builds. A different world, moving at a completely different speed.

Same marketer. Two opposite realities. If you want an honest answer about using AI for marketing, you need to hear from both of them. That’s this article.

Key Takeaways

• AI doesn’t replace marketers. It replaces the mediocre ones and multiplies the best. Your three strongest people with AI produce nearly the same output as a team of ten, at better quality
• Two and a half years in, I do the same job in a completely different way. Tools that were fundamental to me a few years ago have almost disappeared from my workflow
• What it’s genuinely great at: holding technical quality steady, copywriting with a human hand on it, and meetings where nobody can say they didn’t know
• What it still gets wrong: it obeys imprecise instructions literally, it has no sense of sequencing, and it cannot think outside the box. Someone has to do that for it
• The real cost is around 200 euros a month for heavy use. The expensive part is the competence of the person driving
• It’s AI applied to marketing, not marketing applied to AI. Marketing does not orbit the machine

AI Doesn’t Replace Marketers. It Replaces the Mediocre Ones.

The uncomfortable part first. If you are a professional whose output sits below average, say building websites, the machine is already better than you. Not slightly. Clearly. I don’t expect that gap to close.

If you are good, the opposite happens. The machine does not beat you. It removes the ceiling above you, and you operate noticeably above average, on demand, in domain after domain.

Which leads to the multiplier, and this is the core of everything I believe about using AI for marketing:

Take your three most qualified people, give them AI, and you get nearly the same output as a team of ten, at better quality than the ten produced before.

I live the extreme version of that math on my second computer, where I am a marketing department of one. There the multiplier is not a theory I read in a report. It is the reason my calendar works.

One guard rail before we go further, because it runs through everything below: none of this means designing your work around AI. AI is not the center of marketing. It is leverage applied to marketing. I’ll close on that sentence, because it’s the one I would put on a wall.

Same Job, Completely Different Practice

I’ve been using AI radically, meaning daily and in production, for about two and a half years. Not experimenting. Running real work through it.

In that time my process has changed completely. I still do the same things: strategy, content, analysis, websites, campaigns. I do almost none of them the way I did before. Tools that were fundamental to me a few years ago have nearly disappeared from my workflow, replaced by others that simply work better. I rebuilt my entire marketing tech stack for this way of working.

And I have no reason to believe this is the end state. I don’t know whether in two years I’ll be working completely differently again. I expect so. I remember 1998 to 2006, years in which the way we worked kept changing every few months. My honest expectation is that this cycle will be faster than that one, not slower.

That has a practical consequence for you: the specific tools I could name in this article matter less than the operating principles. Tools rotate. Principles hold.

The Sparring Partner Nobody Talks About

The most underrated use of AI in marketing is not output. It is resistance.

I work at a whiteboard. I bring ideas I like, and I want them hammered and torn apart before anyone else sees them. An assistant configured with real context about your business does exactly that. It finds the weak assumption inside the plan you were proud of.

The second thing it catches is the blind spot between specialists. The designer checks the design. The webmaster checks the structure. Neither challenges the other’s domain, so the inconsistencies that live exactly between the two survive every review. A generalist AI holding context on both will flag them in seconds. It starts to feel like a person who is getting to know you and learning alongside you. You are not alone anymore.

It also changes what senior means. The old rule said the higher you climb, the less hands-on you are, and leadership drifted toward admin. With AI the turnaround is so fast that I can be as operational as the team, on real deliverables, in the same week. For a one-person marketing function, that is not a nice-to-have. It is the job.

What It’s Actually Good At

Three areas where the difference is undeniable in production. Not in demos.

Holding technical quality steady. In competitive markets you need to rank first, and you need the technical prerequisites: a fast, clean site. Very few people deliver a site that is beautiful and technically impeccable at the same time. On WordPress, one collaborator uploading one uncompressed image is enough to collapse Core Web Vitals for the whole domain. With AI in the workflow, “upload this image and make sure the Core Web Vitals stay stable” becomes an instruction anyone can follow. The machine compresses, cleans the structure, and keeps the environment fast. Someone without deep technical skills now maintains a production environment that used to require a specialist.

Copywriting with a human hand on it. Left alone, AI writes fast and bland. Interesting and compelling does not come for free. Guided paragraph by paragraph, “this is dull, change this example, you lost me here”, the final result is probably ten times better than what I would produce unaided. The guidance is the work. I wrote about the content-generation side of this in my AI marketing playbook, and the same rule applies: the differentiation still comes from you.

Meetings that close. A recording becomes a summary in about ten minutes, with an owner on every action. The chronic corporate disease of “I didn’t know that” stops being an excuse. Full automation is not there yet: summaries still make mistakes, and I still check them. But the accountability shift is permanent.

What It Still Gets Wrong

Most “AI destroyed my site” stories are human instruction stories. You look at mediocre work and say “this isn’t good, delete everything and start over.” The machine deletes everything. It did exactly what you told it to do. (Unless the site was already a mess, in which case starting over was the right call. But then say that, precisely.)

Sequencing is the deeper limit. AI sees an error and wants to fix it now. On a real project, the correct order is often “leave that for the end, otherwise you will end up rebuilding everything.” Knowing the order is architect work. A good prompt is not enough to hand over the reins of a project, and anyone who tells you otherwise has rarely shipped anything complex.

Then there are the two excesses, and most companies sit in one of them. “It’s not good enough for us” versus “it’s already perfect.” Both are wrong. It is excellent at some things and bad at others, and the expensive skill is knowing which is which. When companies get this wrong, the failure patterns are depressingly repetitive; I documented the four big ones here.

And the risk nobody prices in: atrophy. If you delegate everything and learn nothing, you erode the exact competence that makes you good at driving the machine. The floor keeps rising. Below-average work has never been more replaceable, which is the thesis of this article arriving from the other side.

Why My Employer Can’t Use Any of This Freely

Back to the first computer. Some readers will hear “strict AI policies at a multinational” as backwardness. It isn’t. It is responsibility, and it deserves an honest explanation.

The policies exist because data privacy has to be absolutely guaranteed. Right now, practically no AI company offers that guarantee with 100 percent certainty. We don’t know what happens to data fed into frontier models, and there is a real risk it gets used on behalf of third parties. A multinational holds its own critical information and, above all, its customers’ information. It cannot afford to get this wrong, so it doesn’t gamble.

So inside the company, AI runs where policy allows, and the rest of the leverage stays on my second computer. Two stagnant worlds, completely separate, and maybe that separation is temporary.

My bet on where they meet: local models. Frontier-class models, compressed, running on low-power local hardware instead of a data center that burns water. Models like Kimi K3 and GLM 5.3 are already showing what compression is doing to the quality gap. A company that runs the model on its own metal, with its own data, never leaving the building, does not have to trust anyone’s privacy policy. When compression preserves quality, corporate AI policy stops being a wall.

There is a historical rhyme here. They sold the internet wrong, it bubbled, the bubble burst, and then it quietly became the normal substrate of everything. I expect the same arc for AI, and I would rather be early on the “normal substrate” side than on the “wrong idea” side.

What Using AI for Marketing Actually Costs

The meme says “I fired my staff and now AI costs more than my staff did.” That was true for a brief moment. It isn’t anymore.

Today, heavy professional use of frontier models runs at roughly 200 euros a month. A hyper user pushing every tool to its limit sits at 400 to 600. For that you get, in effect, an unlimited assistant with a decent brain, on call, in every domain you know how to interrogate.

Compare that number to any other line item in your marketing budget. Then notice what the real cost is. The subscription is trivial. The competence of the person driving is the expensive part, because it decides whether the same subscription produces a mediocre site or a production-grade one. That is also why the “AI costs more than people” argument feels stale from both directions: the machine was never the alternative to people. Competent people plus the machine is the actual configuration, and mediocre-without-the-machine is the one disappearing.

How I’d Test Whether Someone Can Actually Use AI for Marketing

Interviews about AI produce rehearsed answers. So when I need to know whether someone can actually drive these tools, this is what I put in front of them: a terminal, access to a good model, three projects. It is also what I would give anyone starting from zero with AI who needs to build the competence fast.

Project one: build a website. Then tell me what’s wrong with it, what you would do differently, what you couldn’t do, and what you think is missing. I’m not grading the site. I’m grading whether you can see past what the machine handed you.

Project two: a market analysis, your way. Then defend it. How did you build it? How would you have wanted to build it? Are the data correct? How sure are you of the accuracy? This is where hallucinated confidence dies.

Project three: free choice. Build something you believe this company needs. Anything.

The first two projects test competence with the machine. The third tests the thing the machine doesn’t have. AI does not think outside the box, so I need someone who does it for the machine.

AI Applied to Marketing, Not the Other Way Around

Put all of this together and you get the sentence I would keep if I had to delete the rest: marketing does not revolve around AI, and it never should. AI is a tool that gives marketing leverage. Used well, the quality, the numbers, and the speed all go up at once.

It’s not marketing applied to AI. It’s AI applied to marketing.

If that distinction sounds obvious, spend a week watching what companies actually do. Too many of them are rebuilding their marketing around whatever the model of the month happens to do well, and then wondering why nothing feels like marketing anymore.

One more thing, honestly labeled: there is still enormous ignorance about AI applied to real business, and I’m building a project for exactly that. A practical course on using AI in business, structured as a small ongoing subscription rather than an expensive box, and in constant evolution, because the ground under it keeps moving. It’s in progress, a waitlist is coming, and nothing about it is final yet. I mention it because its motive is the same motive as this article: too many people are being sold the wrong idea, and competence is the scarce resource.

If you take one action from this article, take project one. Open a terminal, build something, then tell the truth about what’s wrong with it. That loop, repeated for two and a half years, is the whole method.