Generative AI Walked Into the Marketing Department: What Stands Between "Can Write" and "Can Run in Production"

The other day I grabbed dinner with a friend who runs a brand. He pulled out his phone and showed me — his team had just used ChatGPT to write a whole week of posts for their WeChat Official Account (China's dominant business-publishing platform).

I asked him, how do they read?

He laughed. Smoother than the intern I hired.

Then he added: after we published, conversions dropped thirty percent.

That got me thinking for a long while. The story of generative AI in marketing has been told and retold these past two or three years. How it can write copy, draw images, cut videos, personalize recommendations, even make decisions for you. Every one of those claims sounds right.

But sit down and actually ask the question: how does it really get inside a company's marketing pipeline? And once it's running, where does it get stuck?

Not many people can answer that.

First, What It Can Do: From Typist to Every Role

The things generative AI can do in marketing, when you break them open, really come down to two pieces.

The first piece is making things. Words, images, video. ChatGPT, DALL-E, MidJourney, Synthesia — you've probably heard all these names. What they're replacing is a string of "production steps." Drafting the first version of a post, pairing it with a poster, cutting a talking-head video. What used to get done by stacking people now gets done by stacking models.

The second piece is worth more money: making things differently for different people. The same landing page shows a casual style to user A and a business style to user B. The same email has its subject line, imagery, and recommendation slots tuned to each individual. Then you feed the clicks, dwell time, and bounce rates back in, and let the model judge how to adjust the next wave.

Sounds great in theory, right.

But don't rush in just yet. First, look at a few industries where it's actually running — and see how they put it to real use.

A Few Openings That Genuinely Work

Retail and fashion moved first.

Why? Because this industry's content consumption is staggering. A single apparel brand launches hundreds, even thousands of new SKUs in a season, and every one of them needs an image, copy, and a product detail page. People can't write that fast; outsourcers deliver uneven quality. Generative AI is the first time the math actually works on this.

The food and beverage industry took a different route. They use AI for divergent ideation. A new product launch used to need a week of brainstorming to settle on a slogan and packaging concept — now it can churn out dozens of versions in a few hours, and the marketing team picks from the pile.

Travel cut straight to it. AI-generated destination videos, virtual tours, personalized itinerary recommendations are becoming standard practice for customer acquisition. Dwell time and booking conversion rates both show a visible lift.

Spot the pattern?

What makes generative AI genuinely valuable in marketing is that it makes "trying" cheap.

Trying an idea goes from a week to an afternoon. Trying an audience segment goes from a single campaign to something you can swap out anytime. That cheapness is the real reason it's replacing a piece of the cost structure.

But the Sticking Points Are More Than You'd Think

That's the upside. The next sentence is the key.

What's genuinely hard isn't getting AI to write something — it's getting the company to dare to use it, to be able to manage it, and to keep it from blowing up in your face.

I'll break it into five hurdles.

Data privacy. Personalization has to be fed user data — how much, is it compliant, how do cross-border transfers get handled? A single GDPR fine landing, and several million euros are gone.

Ethical risk. An AI-generated image shows up with something that shouldn't be in the frame, the copy offends a particular community, a recommendation algorithm gets called out for discrimination. There have been several such incidents in 2024 and 2025, big and small.

Employee resistance. This one is the most easily underestimated. Copywriters, designers, video editors — their first reaction is usually "AI is going to replace me," not "AI is here to help me." Once that psychology spreads, the rollout stalls.

Quality control. The copy AI writes sometimes reads smoothly but gets the facts wrong, drifts off the brand's tone, or even directly lifts a competitor's sentence. Without a strict review gate, you're just digging yourself into a hole.

Infrastructure — the most pragmatic one of all. To plug AI into the marketing system you need a data middle platform (a centralized data layer shared across business apps), a content review workflow, version management, and an A/B testing loop. Plenty of companies haven't even sorted out their content asset library — what business do they have talking about AI?

The Plain Truth: the Gap Is in the Organization, Not the Tool

The tools are off-the-shelf. ChatGPT, MidJourney — anyone can use them.

But the same tool turns Company A into a growth engine and Company B into a PR disaster. Where's the gap?

The gap is in the "people + process + data" foundation.

A company with a solid foundation treats AI as an amplifier — it takes what's already working and magnifies it tenfold. A company with a broken foundation gets AI as a magnifying glass — it takes the existing chaos and magnifies that too.

So the advice I gave that friend — the one whose conversions had dropped — was simple. Don't start by fixating on how smoothly AI writes. First ask: does your brand have a clear "brand voice guideline" for AI to align to? Does your content review workflow have a hard gate where a human signs off? Has your team come to treat AI as a collaborator rather than a replacement?

Answer those three questions, and only then can AI actually move from "can write" to "can run in production."

Marketing was never about who writes faster.

It's about who understands people better.

AI has learned to write. But it hasn't yet learned to understand people.

The distance in between is the homework those of us in marketing genuinely have to do over the next two or three years.