Before You Market with AI, Settle These Three Accounts First

A while back, an old classmate of mine who works in marketing invited me to dinner. He's in the B2G (business-to-government) business, helping companies chase government projects: writing proposals, producing materials, running tenders.

Halfway through the meal, he pulled out his phone to show me the new "colleague" his team had taken on: ChatGPT writes their social media copy; Midjourney makes the images; even the first draft of his reporting decks starts as an AI-built frame that he edits down.

He was quite pleased with himself: work that used to take a month now gets done in a week.

I said, sounds great.

Then he sighed. Last Friday, just before end of day, an AI-written tweet nearly went out as-is. Some instinct he couldn't name made him take one more look—and he found it cited a "statistic" that led nowhere.

"So now I'm torn," he said. "This thing—how much of it should I actually trust?"

It's a good question. Good enough that I drained what was left in his glass and thought about it the whole way home.

AI has walked through marketing's door, and that door isn't closing again. But once it's inside, you'd better know how to run the numbers. I've tallied up three accounts. Here they are.

First, Let's Get One Term Straight

What is generative AI, exactly?

Start by picturing the old AI as an accountant. You hand it a pile of data, and it counts for you: which channels convert best, which kinds of users click which links. It produces reports; it doesn't act.

Generative AI is the one that acts on your behalf.

Tell it "write me a WeChat Moments post for the new product," and it whips out ten options, then asks which one you like. Say "give me a poster," and tools like OpenAI's DALL-E and Midjourney really will paint one in seconds, in whatever style you pick. Google's Gemini goes further still: it can search the web while it writes, then turn the finished draft into several languages on the side.

Writing, drawing, voiceover, video editing. Nearly everything on a marketer's desk, it can lend a hand with.

Good tool, no question. So what's the price?

The First Account: The Money It Saves You

Start with the gains—hard numbers first.

Do the math on a social media manager grinding out ten posts a day. At half an hour per post, five hours are gone—and that's before counting the images. Images are either outsourced, at a few dozen to a hundred-plus yuan a pop, or you queue behind your in-house designer and land next Wednesday.

And now? AI turns out ten posts in minutes, images included. Product descriptions, marketing emails, event announcements—nearly all of that standardized work can be handed over.

And the people you free up? Set strategy. Meet clients. Spend the brainpower you just reclaimed on the work that truly demands big ideas.

And that's only what's on the surface.

Take personalization. What AI does best is "read people": it goes through users' behavior, preferences, and purchase records with a fine-tooth comb, sorts them into a stack of segments, and hands each segment the one line that lands right in their heart. This kind of delicate work used to take a data analytics team on staff, grinding away slowly. Now, it's a switch.

Then look at service. Among your customers are night owls, early birds, and someone in Australia—where it's mid-afternoon while it's 3 a.m. on your side, and he needs an answer now. Who picks up?

Staff a night-shift support team? Run that cost math yourself.

Most companies' answer these days is the chatbot. Working all year round, endlessly patient, same even temper while a customer lays into it at 3 a.m. Whatever it can solve, it solves on the spot; whatever it can't, it hands to a human. Your experts end up handling only the genuinely hard problems, no longer walking people through "have you tried turning it off and on again" for the tenth time.

Two more items sit further back, but they carry real weight.

One is creativity. Whether AI truly counts as creative is still being argued across the industry. But some of the connections it throws up in a brainstorm genuinely wouldn't have crossed your mind. Use it as a whetstone to force your team to think wilder; take what it produces as a draft, tweak and adjust, and make it your own.

The other is decisions. AI doesn't just count; it can "read" the numbers too. Its predictive analytics can help you gauge trends and spot opportunities, telling you where to double down next quarter; keyword bids and selections can be tuned in real time against live performance.

Efficiency, personalization, round-the-clock coverage, creativity, decisions. Five line items, every one of them a gain.

But.

The Second Account: What It Might Cost You

Everything has a flip side. Those five words are worth taping to the edge of your monitor.

First, the copyright muddle.

Generative AI "learned" from material scraped across the whole web. So the text and images it produces aren't necessarily original—they may even closely resemble certain original works. Who infringed? Who pays whom? Even the foundational question—whether AI-generated output counts as a "work" at all, and who owns the copyright—is still being chewed over by legal systems around the world.

Wherever the law hasn't made up its mind is exactly where your risk exposure runs deepest.

Second, the muddle over what's true.

There's an old saying: ninety percent of what's online can't be trusted. Overstated, but the logic holds. AI feeds on exactly that online information—it reads a lot, but it can't tell true from false. It serves up the wrong as if it were right, and with total confidence.

That's exactly where my classmate's near-miss tweet came from. Anyone whose business is government work will feel this in their bones: one wrong figure in your bid materials, and the whole bid can sink.

Third, the muddle over where the line is.

AI read the whole web, and it absorbed the whole web's biases along with it. It can produce content that offends particular ethnic groups or faiths, traffic in stereotypes, even end up vouching for extremist views.

And the deadliest part? It isn't doing any of it on purpose. There's no malice, because there's no "intent" there at all.

One more for the list: privacy. AI has to swallow oceans of data to do its job, and some of that data is users' private business. What may be touched and what must be left alone—the model genuinely can't tell. Europe has GDPR, California has CCPA; step over the line and you're looking at fines and lawsuits, both without fail.

Notice what copyright, truth, line-drawing, and privacy have in common: AI doesn't know which money can't be made, which words can't be said, which data can't be touched.

And these four happen to be precisely the ones that can cost you serious money.

The Third Account: Where Humans Must Stand

Some will say: just set up a few more AIs—what do you still need people for?

That brings us to AI's hardest limitation: it has no human touch.

It can write a perfectly steady financial summary, but don't expect it to write copy that makes a heart stir. It imitates emotion; it doesn't understand emotion—and readers can feel the difference. For financial reports, for technical documentation, it's passable. For marketing, the moment the job becomes "move people," it has hit the end of its road.

So the smart way to use it: AI paves the road, humans walk the final stretch. Let it lay down drafts, build frameworks, produce raw material—and let your team polish on top of that base, working in the emotion, the warmth, the connection with real people, bit by bit.

Quality works the same way. Copy, emails, scripts—it can produce them all, but the output can't go straight out the door. It's a half-finished product, and the value of a half-finished product takes a human to cash in.

And the jobs? That worry hasn't stopped since the day AI blew up. The most directly hit are roles like graphic design and image generation. The optimists say: don't panic, the roles will change shape—image people shift into AI management, feeding prompts, running experiments, reviewing results, then iterating.

My own judgment: this job shift is real, not a consolation prize. A team genuinely does need people who can push AI to its limits—know how to experiment, how to evaluate, and how to put it to work. But that only works if you're first willing to spend the money to train them. Skimp on that training, and the money you saved up front will find its way back out of your pocket, sooner or later, in some other form.

Put plainly:

Efficiency comes from AI. Judgment has to come from you.

Back to That Dinner

Later I sent my old classmate a WeChat message. One line: treat AI like the most capable intern you've ever had.

An intern works fast, never sleeps, and comes cheap—but what they hand in, would you dare publish it without so much as a glance?

You wouldn't. So keep one eye on it.

AI handles fast; you handle right. AI handles volume; you handle heart. The day something truly goes wrong, the signature is yours—and so is the accountability.

Here's wishing you many more hours saved by AI—and wishing you never once have to get up at 3 a.m. to clean up the mess it made.