Marketer, Is Your Job Still Safe?

A friend who runs e-commerce was venting to me the other day.

He said his company used to keep eight copywriters on staff, just to write product detail pages. Now? They kept two. The work of the other six got handed to a single tool.

What eight people used to do, two people plus a machine now handle.

I asked him where those six copywriters went. He smiled and didn't answer.

Something caught in my chest. Not for those six — for everyone still doing marketing the old way.

From "Guess What You Like" to "We'll Write It for You"

Think about it. For the past decade, when marketers talked about AI, what were they actually talking about?

Recommendation systems. Hyper-personalization — what Chinese e-commerce calls "a thousand faces for a thousand people." "Guess What You Like" (the ubiquitous recommendation module on every shopping app).

Plainly put, it was all about picking from what already existed. The algorithm sifted through a pile of products, fished out the one most likely to make you pay, and pushed it in front of you. The big e-commerce platforms had figured this game out long ago.

But over the past year or two, the wind shifted.

The new wave of tools doesn't pick anymore. It creates from scratch.

Product detail pages — it writes them. Promotional posters — it designs them. Short-video scripts — it drafts them. Customer-service replies — it answers them. What used to take a whole creative team a week to polish now rolls out in minutes. Fifty versions at once, if you want.

From "helping you choose" to "doing it for you" — that's a qualitative leap.

So, How Do You Actually Use It?

Here's where it gets thorny.

I've talked to several friends in marketing, and I keep seeing the same problem: the tools are bought, the accounts are opened, but nobody can make them feel like they're working. Today they have the machine write a couple of ad slogans. Tomorrow they generate a few images. Scattershot. Hit-or-miss.

All firepower, no battle plan.

Worse, a lot of teams haven't even figured out "what problem do I actually want AI to solve for me" before they rush in. Three months of flailing later, the boss asks how it's going — and nobody can answer.

I think this needs a playbook. Not the hand-wavy "five-step digital transformation" kind — a real, on-the-ground, step-by-step route.

A Route That Actually Works

I've broken it into seven steps. Hear me out, then judge for yourself.

Step one: figure out what you're actually trying to do.

Sounds obvious, but some companies genuinely skip this. You have to answer one question: with AI, are you trying to save money, win more customers, or build your brand? Different goals, completely different paths. An e-commerce company trying to take the lower-tier-city market (China's tier-3-and-below cities) and a bank trying to hold on to high-net-worth clients — can they really play the same game?

Step two: take stock of what you've got.

AI grows up eating data. What data do you have on hand? Customers' purchase records, browsing trails, service-chat logs, social-media comments. And don't just stare at your own stuff — outside public data and industry trends can all be fed in. A friend at a food-delivery platform told me they even use data like "which city searches for which dish most at which hour," and the ad copy it generates is so locally authentic that even locals find it warm.

Step three: clean up the data.

This is the most tedious step, and the most critical. Garbage data in, garbage out. Especially in multilingual environments — just smoothing out dialect and colloquial expressions can give a team headaches for half a month. But somebody has to do the dirty work.

Step four: pick your tools.

There's too much to choose from on the market right now. One category writes text, another makes images, yet another generates video. Don't get greedy — pick what fits your business closest. If you sell clothing, nailing virtual try-on and product shots first matters more than anything else.

Step five: tune it.

Think you can just use a general model off the shelf? You can — but it won't be good enough. You have to feed it your own historical material. One FMCG (fast-moving consumer goods) giant dug out decades of its own ad archives to train the model, and the region-specific ads it produced felt more down-to-earth than the expensive work they used to hire 4A agencies (top-tier multinational ad agencies) to make.

Step six: inspect the goods.

What the machine writes can't go straight out the door. You need to check a few things: Are the facts right? Did it hallucinate? Could it offend anyone? And once it's out there — did click-through rate and conversion rate actually go up? One bank ran a comparison test: the AI-generated personalized campaigns had a click-through rate a quarter higher than the human-written version. That's the number that counts.

Step seven: roll it out.

Once it's validated, roll it out boldly. Ad slots, social media, product pages, customer-service windows — plug it in everywhere you can. And then the most important thing of all — watch the feedback, keep tuning. This isn't a one-shot deal.

Some People Are Already Making It Work

Let me tell you a few true stories.

One e-commerce giant had been trying to break into tier-3 and tier-4 cities for a while, but the ad copy always had a stiff, out-of-towner tone. Then they used AI to mass-generate dialect versions of their ads, tailored to local festivals and customs. That year, during the big sales festival, click-through rate rose nearly 20%, and creative-production costs were cut by 30%.

One FMCG giant had a headache: India is huge, with hundreds of languages and dialects — impossible to shoot separate ads for every region. They used AI to generate different-language, different-cultural versions of ads for the same product. Five versions a quarter before; now fifty versions a month.

One travel platform plugged in an AI itinerary planner. You type "taking my parents on a trip south, five days, budget five thousand," and it lays out a decent plan. User-satisfaction scores climbed more than 10%.

One bank, in the most tightly regulated lane of all — finance — somehow still managed to nail personalization. They anonymized the data before feeding it to the model, and the credit-card campaigns it generated felt spot-on without crossing any red lines. Click-through rate, again, up a quarter.

Notice — e-commerce, FMCG, travel, banking. Four completely different industries, all producing visible results. Proof that this playbook isn't any one industry's monopoly.

But

Don't celebrate just yet.

Wherever there's a good thing, trouble follows.

First, the data hurdle is hard to clear. Big companies have stockpiles; small companies? The customer data they're holding might be thinner than what's in a WeChat group. Without material, no matter how smart the machine is, it can't pull a rabbit out of a hat.

Second, trust. This one's delicate. Users want you to understand them and show them the right things — but they're also afraid you understand them too well and feel surveilled. One e-commerce company's personalized ad accidentally leaked a user's exact purchase history, and the user blew up on the spot. Personalization pushed too far becomes a jump scare.

Third, cost. Training and running models burns real money. Big companies grit their teeth and get through it; small companies might not even afford the entry ticket. Which creates an awkward situation: the small and mid-sized businesses that need AI efficiency the most are exactly the ones who can't afford it.

Fourth, the rules. Every jurisdiction is tightening the screws on data compliance. The premise of personalization is getting user consent; generated content can't be faked and can't carry bias. Step over the line, and the fine isn't small change.

At the End of the Day

After that conversation with my friend, on the way home, I couldn't stop thinking about one thing.

For the marketing industry, what does this tool called AI really mean?

A cost-saving machine? An efficiency lever? Or the beginning of the end for humans?

I turned it over and over, and I don't think it's any of those.

It's a watershed.

The people who know how to use it, the teams who know how to think with it, turn it into an amplifier — amplifying creativity, amplifying efficiency, amplifying their understanding of users. The ones who don't are still stuck on "does machine-written copy have a soul?" — and by the time they look up, the market is already gone.

Those six copywriters who left my friend's company weren't eliminated by AI. They were eliminated by two copywriters who knew how to use AI.

Tools never eliminate people. People who can wield new tools eliminate people who can't.

That truth holds in any industry.

And as for that user, scrolling through their phone late at night, suddenly struck by an ad that feels "too spot-on" — what they don't know is that no human copywriter may have touched that ad at all. They just feel: this brand gets me.

What gets them isn't a person. It's data and an algorithm.

Linger on that too long, and it sends a chill down your spine. But then again — who cares? The product works, the ad looks good, and that's enough.

Maybe someday in the future, we'll look back fondly on that era when ads were clumsy but unmistakably "written by a person."

Maybe we won't.