AI Can Write Your Copy. It Can't Make Friends for You.

A few days ago, a friend of mine who works in marketing showed me something.

He turned on his computer and typed in one line. A few seconds later, hundreds of lines of ad copy came pouring onto the screen — headlines, selling points, calls to action, the whole package. He typed another line, and out came hundreds of design drafts.

He turned to me, a little smug: "Fast, right?"

Fast. Really fast.

I said, I have a question for you too: this copy — would you dare send it to your most important client, not changing a single word?

He froze for a moment and had nothing to say.

That frozen moment is what I want to talk about today.

The Speed Is Real

First, let's be clear about what generative AI actually is.

The automation we used to talk about ran a fixed process faster. Generative AI is different: it can "grow" content of its own — copy, images, video scripts, product descriptions, pricing suggestions, even drafts of customer-email replies, generated on the spot.

By 2025, these platforms had already run through every one of these jobs.

What can marketing teams do with it? Test hundreds of creative versions at once. Give millions of users one tailored line each. Preview customer reactions before ever running a single field study.

It can handle lead mining, first drafts of proposals, technical documentation too.

Powerful. Truly powerful.

But.

Marketing isn't like other trades. It's half craft, half relationship. It builds trust. It conveys "this is who I am." It shapes how people will see you for years to come.

And these are precisely the things algorithms are worst at.

At its core, generative AI is statistical inference. Think of it as someone who has memorized every cookbook but never set foot in a kitchen. It has no sense of context, no emotion — let alone ethical judgment. What it says depends on what you feed it, and what you tell it to optimize.

In another department, that might be a minor flaw. In marketing, it's a big problem.

Because it stands between you and your customers.

Let me walk through four scenarios. One at a time.

Scenario One: It Knows What You Bought, Not What You Went Through This Week

Personalization is the most profitable use of generative AI in marketing.

The system stitches together your browsing history, purchase records, audience tags, and third-party data, then rewrites headlines, swaps offers, and tunes recommendations in real time. The results are immediate: open rates up, clicks up, conversions up.

But what it optimizes is "the statistically most likely option" — not you, the person.

Think about it. This week you looked up an illness for an elderly relative. For the next month, every time you open your phone, the screen is full of recommendations related to it. Each one reminds you of something you would rather not be reminded of, over and over.

It's not malicious. It just doesn't know.

It has no sense of tact. And tact is the most expensive thing in any human relationship.

Customers feel offended, feel manipulated — they screenshot it and complain in public. One ill-timed push notification, and half the goodwill a brand spent years building drains away.

The algorithm can predict what you might buy. It cannot predict whether you can bear being interrupted right now.

Inside companies, things are shifting too. The copywriter's job has changed from "writing" to "reviewing." Hundreds of machine drafts a day — efficiency is up, but the work is no longer his own. Guess whether he still wants to talk about his job when he gets home at night.

Scenario Two: You're Not Asking Customers — You're Asking a Mirror

Market research used to mean surveys, focus groups, field observation. Expensive, slow — but with texture.

Now there's a new toy: synthetic consumers. Simulate a crowd of "customers" with virtual personas — ask them whether they're price-sensitive, which flavors they favor, which line hits home. Instant answers.

At first, the answers matched the historical data quite well. The boss did the math: huge savings. The traditional research budget? Cut.

But. What is a synthetic consumer? The average of historical data. A pattern aggregated out of the past — not a living person.

It doesn't know the meme that popped up this week. It doesn't know why a niche community suddenly blew up. And it has no idea that young people's shared context turns over every three months.

You think you're listening to your customers. You're listening to the echo of your own database.

However sharp the mirror, it cannot show you tomorrow.

So here's my advice: let AI propose the hypotheses, and let humans verify them. It can help you get the questions out. The answers you still have to find on real customers.

Scenario Three: What Luxury Fears Most Is a Mass-Produced Soul

In creative design, the efficiency gains are even more startling. Feed in a text description, and out come hundreds of design drafts in minutes.

Mass-market brands are delighted: the cost of trial and error drops to nearly zero — experiment freely.

Luxury brands, on the whole, are hesitating.

Why? Because the value of luxury is built on scarcity and craftsmanship. One bag takes a master artisan dozens of hours to stitch. One watch passes through more than a hundred finishing steps. What the customer buys is not just the object, but the story that someone spent time on it.

The images AI produces are beautiful. Truly beautiful.

But they have no provenance, no human warmth. One glance and customers see a machine drew them — and then they ask: does that craftsmanship story you've been telling for decades still count?

Inside, there's anxiety too. Designers have split into two camps: one treats AI as an inspiration library and runs with it; the other believes the craft that feeds their family is about to be replaced.

And anxiety is contagious.

Scenario Four: When It's Wrong, It Doesn't Stumble

Now look at B2B.

Much of the marketing content in this field is technical specifications, compliance documents, engineering instructions. One wrong number here can mean an accident on site — even legal liability.

Generative AI writes this documentation at a pace where one person matches a small team. More leads, less headcount, and the reports look great right away.

But it has a flaw: it will talk nonsense with a perfectly straight face.

When it's wrong, it shows no guilt. However neatly typed, wrong is still wrong.

Without experts at the gate, errors sail straight through to the client. Let enough small mistakes pile up, and trust built over decades can be gambled away in a few months.

And there's a more hidden loss. Lay off the experienced veterans, and you throw the organization's institutional knowledge away with them. Sales loses the sounding board it used to bat plans around with. The company starts to depend on a system it can no longer explain.

What you save is wages. What you pay out is judgment.

When Roles Change, the Rules Must Change Too

Push these scenarios forward, and you arrive at the organization itself.

Roles slide from "creating" to "supervising." Machine-generated content goes out; something goes wrong — who owns it? It used to be crystal clear. Now it's a muddled mess.

Go a level deeper: How should user data be used? Is the algorithm biased? Who owns the copyright? Should you tell the client "an AI wrote this"? When false information gets out, who takes the blame?

Deploying systems without rules is like merging onto the highway without fastening your seatbelt.

Then there's the human problem. When the machines do all the work, all that's left for people is clicking "confirm." Nobody likes a job of clicking confirm. First their minds wander. Then they walk out the door.

How Humans and Machines Team Up

So what now? Give it up?

Use it. Not using it is picking a fight with yourself.

But there's a right way to use it, and it has a name: the hybrid model. In plain words — humans are the brain, machines are the hands and feet.

The brand's voice, the boundaries of what it says: humans decide. Whether the technical content is accurate: humans verify. Cultural signals, context, tact: humans read them. And the heavy lifting — scale, testing, optimization — goes to the machines.

To actually put this in place, do at least five things:

1. Whatever the machine produces gets a human review before it goes out the door;
2. Keep an escalation path open for sensitive content — never let the machine make the final call;
3. Audit AI output on a regular schedule, and trace back whatever went wrong;
4. Pull together a cross-functional governance group — don't let the tech department call all the shots;
5. Keep training your people, so human skill grows alongside the machines.

Before you go live, take an inventory: the repetitive work, the pure optimization, the first drafts — hand them to AI. The work that calls for judgment, for empathy, for years of accumulated expertise — keep it in human hands.

And be straight with your employees: how the roles will change, where the road leads. Being cagey about it only invites more resistance.

The most important one: treat it as a strategic asset, not a cost-cutting tool.

The money you save on headcount will never prop up a brand.

Back to That Question

As I was leaving that day, my friend asked me: so tell me — should we use AI or not?

Use it. Of course.

But remember: efficiency is its job; trust is yours. It handles the fast. You handle the true.

Marketing, in the end, is a relationship between people. Brands live in emotion, live in culture, live in the way people look at one another. And these are the things an algorithm cannot fully compute.

The future of marketing is not artificial. It's collaborative.

And one more wish for you: use AI as an amplifier, not a crutch.