Generative AI Has Entered the Marketing Department: Blessing or Trouble?

A little while ago, a friend who works in marketing at a Fortune 500 company started venting to me.

His company had just plugged the entire marketing department into generative AI. Work that used to take a copywriting team two weeks was now coming out in half a day. The boss was thrilled. The CFO was thrilled.

He wasn't thrilled.

"Drafts I produce with AI — even I can't tell anymore whether I wrote them or the machine did. And on the client side? Engagement is up, but the feedback is full of 'fake,' 'empty,' 'no warmth.'"

He asked me: "Is this thing a blessing or a curse?"

I thought for a moment, and gave him an answer he didn't much like.

It's not a blessing, and it's not a curse. It's an amplifier.

Use it to build something good, it amplifies the good. Use it to slack off, it amplifies the slacking. It amplifies your company's capabilities — and it amplifies your company's biases.

Today I'm going to break this down carefully.

I. First, let's get clear: what are companies actually doing with it?

When ChatGPT landed at the end of 2022, the whole business world went into a frenzy.

Inside two years, Gemini, Copilot, Claude, ERNIE Bot (Baidu), Tongyi (Alibaba) — all of them had arrived. When we used to say "AI," what came to mind was "recommendation algorithms" and "precise ad targeting." Not anymore. Now it writes, it draws, it edits video, it codes, it chats with your customers.

Let me list a few real cases, so you can see how deep it has penetrated.

Coca-Cola ran a campaign inviting users to co-create with its AI. In 11 days, users generated 120,000 images, with zero ad spend, and the average person spent 8 minutes inside it.

JetBlue rebuilt its customer-service chat with AI. Every conversation saved 280 seconds, and in a single quarter it saved 73,000 hours of human labor.

Cadbury partnered with an advertising giant in India, took the assets of one Bollywood star, and used AI to cut 130,000 different ads aimed at small shops in different neighborhoods. Final tally: 94 million views. And the budget had been slashed.

Think about that. 130,000 ads. If humans had to do it, how many people would it take? How many years?

That is what generative AI looks like once it walks into the marketing department.

It isn't just a "tool upgrade" — it turned "one-to-one marketing" from an ideal into reality.

II. But it's not that simple

It's easy to let these beautiful numbers go to our heads.

But think one layer deeper: the harder a company leans on AI, the bigger its impact on society as a whole. This is no longer just the company's business.

There's a chain in my head, and it links like this.

First the company uses AI (action). Out of that grow new marketing capabilities (data-driven, predictive, contextual, augmented, agile). Those capabilities then pry the business into transformation (understanding users better, redesigning products, redesigning the marketing mix, deepening engagement, going digital end-to-end). And finally those transformations seep out into society — affecting how you spend money, how you learn, how you deal with other people, how you see yourself.

Action → Capability → Transformation → Impact.

Every step has two sides. Every step has winners and losers.

Let me zoom in on the "action" link first, and look at what's hiding inside the AI models companies are using.

III. Four kinds of models, four personalities

The ones companies use the most fall roughly into four types: the kind that writes, the kind that sees images, the kind that fakes faces, and the kind that recognizes things.

Let's start with the writers. Behind them are Large Language Models (LLMs), with ChatGPT as the standard-bearer.

Where are they strong? They're good at mimicking a human feel for language — writing emails, drafting copy, translating, writing code. They have read more text than you will in your entire lifetime.

But they have a fatal flaw: they can't tell true from false. Without you noticing, they will confidently make things up. The academic term is "hallucination." They also reproduce the biases in their training data — gender, race, politics, all of it. OpenAI itself has admitted its models lean toward a Western perspective.

Next, the image-seers — Large Vision Models (LVMs). DALL·E, Midjourney, and similar image-generation tools all rely on them.

They can recognize, classify, and generate. Medical imaging, autonomous driving, product search — they all depend on them. But they have their own traps: the images they generate may be ripping off someone else's work, the judgment process is opaque to humans (the so-called black box), and training them burns through staggering amounts of electricity.

The third kind fakes faces — Generative Adversarial Networks (GANs).

There's a website called "This Person Does Not Exist." Every time you refresh, it generates a face of a person who doesn't exist at all. The person in the photo looks lifelike — but never lived a single day.

Sounds cool?

Now flip it over. Some people use it to make deepfake videos — forging a politician's speech, fabricating a celebrity's statement, faking your boss on a video call. Privacy, copyright, election manipulation, identity theft — they all come riding in.

The fourth kind recognizes things — Convolutional Neural Networks (CNNs). They're good at finding patterns in images and video. Facial recognition, loan approvals, medical diagnoses — they're in all of it.

The problem is, if the training data carries bias, the model will discriminate against certain ethnic groups when approving loans, or be inaccurate for certain populations when diagnosing disease. And once it makes a mistake, it's very hard to figure out which step went wrong.

IV. Put these four models together — what does the company grow?

Capabilities. New marketing capabilities.

Let me walk through five you should care about most.

The first is data-driven marketing.

In the past, when you looked at user segments, you'd see something coarse like "female white-collar workers aged 25–35." Now AI can take each person's purchase history, browsing trail, social posts, and location, mash them all together, and draw a portrait that only you can read. Then it pushes different messaging, different offers, different imagery to each person.

The second is predictive marketing.

It doesn't only look at "who you are" — it looks at "what you'll do next." Predicting whether you'll churn next week, whether you'll repurchase next month, whether you'll upgrade in six months. Get the prediction right, and the marketing action gets precise.

The third is contextual marketing.

The right message, the right moment, the right setting. Audio on your commute, family content on the weekend, late-night healing copy at midnight.

The fourth is augmented marketing.

AI turns "impossible missions" into "one person can handle it." A copywriter can produce 50 versions at once; a designer can sketch 20 sets of concepts simultaneously. It multiplies everyone's output.

The fifth is agile marketing.

A campaign used to take three months to prep, and once it went live you couldn't change it. Now with AI in the loop, you spot a trend today, ship creative tomorrow, read the data the day after, and iterate.

It all sounds beautiful, right?

I need to interrupt here.

These capabilities didn't fall out of the sky. They want data, they want compute, they want people who know what they're doing. Whether a company has the money to buy GPUs, whether it has clean enough data to feed the models, whether it has people who understand both the business and AI — these determine whether it actually picks up the ticket.

Unequal resources, unequal capabilities, unequal outcomes.

I've talked to executives at large companies, and the thing they say most often isn't "how strong AI is." It's "we spent enormous effort just to get our data clean enough for AI to use."

That is the reality.

V. Once capabilities stack up, the company starts to morph

Once capabilities stack high enough, the company isn't the same company anymore. It transforms in several places at once.

It understands users better. No more surveys, no more focus groups — instead, it mines the needs users themselves can't articulate straight out of vast streams of behavioral data.

It also reacquaints itself. If AI can do all this, then what exactly is the core moat? Is it data? Brand? Channels? A lot of companies start reshuffling at this point.

It rebuilds the marketing mix. Products designed with AI (Toyota using AI to generate EV prototype sketches is one example), pricing adjusted by AI (dynamic pricing, personalized coupons), channels chosen by AI (where to open stores, where to distribute), promotion created by AI (machine-generated posters, scripts, video).

It rewrites customer interaction. Smart customer service online 24/7, shopping bots that remember all your preferences, sales assistants that never tire.

It also rebuilds the digital strategy. From "build a website, open a social account" up to "end-to-end data-driven."

By this point, the company has been fundamentally remade.

VI. And then? And then, society

This is the part that weighs heaviest on me.

When companies change, society is pulled along to change with them. We can't just tally up the company's books and ignore society's books.

The benefits generative AI brings to society — I can count five.

The first is access to wealth. The financial advisor that only rich people could afford is now something AI gives you. It does your budgeting, your allocation, your risk assessment, and it's more reliable than your cousin's stock tips. Some tools have already turned the "robo-advisor" from a luxury into an everyday item.

The second is social inclusion. Visually impaired people can use AI apps to "see" the scene in front of them; hearing-impaired people can use real-time transcription to "hear clearly" what's said in meetings; language is no longer a wall between you and information. Refugees, migrants, people in remote areas — for the first time, they can get the same information as people in tier-one cities.

The third is lifelong learning. It's like a private tutor, building a learning path tailored to you, adjusting difficulty in real time based on your progress. It has turned "one-on-one with a master teacher" from a luxury into a baseline.

The fourth is a smarter everyday life. Smart homes, smart cities, personalized health plans, disaster early warning, energy optimization — AI has let infrastructure learn to "think."

The fifth is personal health. Emotional-companion bots are online 24/7, monitoring the stress signals on your wrist, reminding you to meditate, writing you a personalized meal plan.

Sounds like utopia?

I deliberately said all the nice parts first.

What follows is what I actually want you to be wary of.

VII. Five traps AI digs for society

The first trap: you can't do without it anymore.

You start telling AI your worries, venting to it, letting it make decisions for you. Your contact with real people shrinks. Its answers are always instant, always polite, always agreeable. But it can't give you real empathy — it can't give you that "I see you" look.

Over time, you end up lonelier, more anxious, less sure how to be around real people.

The second trap: your patience is stolen.

You get used to AI replying in seconds. Then, faced with real humans — your parents, your partner, the people you work with — they start to feel slow, annoying, not paying attention.

Relationships start to fray.

The third trap: your mind starts to atrophy.

Every question, you ask AI first. AI gives you the answer, and you stop analyzing, stop questioning, stop digging. You lose the most crucial ability of all — the ability to define the problem.

What's scarier is kids. From primary school on, they're using AI to do homework and look up answers, never going through the process of "getting stuck, struggling, breaking through." The by-product of that process is, precisely, real thinking ability.

The fourth trap: your creativity gets quietly flattened.

The "inspiration" AI gives you is really just the average of its training data. The more you use it, the more your output looks like everyone else's. Everyone is using the same tools, and in the end everyone's work converges.

Personalization has, perversely, become the new homogenization.

The fifth trap: you get locked inside an echo chamber.

AI figures out what you want to see, and shows you exactly that. You think the world is like that — but you're only seeing the version the algorithm wants you to see. Your biases get reinforced. Your grasp of the outside world gets distorted.

And all of it happens without a sound.

VIII. So, policy has to step on stage

Companies will not police themselves. Markets will not self-correct.

This needs rules.

At the individual level, legislation has to grip two things: how deepfake content is labeled and prohibited, and who actually owns user data, who can use it, and how liability is traced when something goes wrong. You can't wait until users have been deceived by a deepfake video to play catch-up.

At the company level, three things have to be made clear: whom AI is replacing, who is responsible when AI errs, and how the original authors of training data should be compensated. Transparency, accountability, retraining programs — none can be missing.

At the national level, two hard questions have to be answered: how do we stop AI's dividends from concentrating in a handful of giants, and how do we prevent it from being used to manipulate elections, fabricate public opinion, and harm national security?

These aren't questions any one government, any one company, any one scholar can answer alone.

It needs people to sit down together. Government, business, academia, civil society — all at the table.

In closing

After I finished talking with that friend, he was quiet for a moment, then said:

"So do I keep using this AI or not?"

Of course you use it.

Don't, and the next wave will leave you behind.

But using it doesn't mean using it with your eyes shut.

You have to understand: what you've plugged into isn't just a tool. You've plugged in a whole chain — capability, transformation, impact. And that chain eventually lands on a real person's shoulders. It could be your customer. It could be your employee. It could be your own kid. Or it could be you, ten years from now.

I love this line:

"Every gift of fate is secretly marked with a price."

What generative AI has brought to marketing, to business, to society — that is an enormous gift.

But the price tag is already on it.

What we can do, before the bill comes due, is read that tag clearly first.