2025: The Year AI Flipped Marketing

A while back, I had dinner with some friends on the brand side.

A few drinks in, one of them pulled out his phone and showed me a number. I did a double take.

In 2025, more than half of advertisers are already using AI to make content.

This isn't a couple of pioneers testing the water — 53.1% of peers are already on it. If you haven't moved yet, you're not being "cautious." You're just slow.

I chewed on this for days. Went back through the campaigns that blew up last year, and the more I looked, the clearer it got: this isn't a technology upgrade. The underlying logic has changed.

What Even Is "AI Marketing" Nowadays?

For a lot of people, AI marketing still equals "have ChatGPT write me a couple lines of copy, have Midjourney spit out an image."

That was 2023.

By 2025, AI has gone from being a "tool" to being an "actor."

What do I mean?

Let me tell you three stories.

Story One: By-Health's Antarctica Film

Last December, By-Health (汤臣倍健) put out a film for its 30th anniversary.

For a 30th, the thinking went, you really should go shoot on location in Antarctica and max out the prestige.

But shooting in Antarctica? Just the location fee alone runs north of a million RMB a day — and that's before you factor in crew, gear, and risk.

So what did they do?

They decided to generate the entire film, 100%, with AI.

The glaciers, the penguins, the expressions on the explorers' faces — even the frost that forms on eyelashes when you breathe — all of it was AI-drawn.

The detail that got me was what came after. To strip out the AI look (that unmistakable AI-feel), they trained the AI the way you'd train a live-action crew. The micro-expressions on every face, the way finger joints moved — they reworked each frame hundreds of times.

In the end: 419,000 social interactions, NSR (Net Sentiment Ratio — share-of-voice positivity) of 44%.

And that's Play One for AI: treat it as a production engine.

It solves a problem that's "too expensive, too hard, or simply impossible to make."

Yili Jinlingguan did something similar: a six-episode animated short series, originally planned for six months. With AI in the loop, they delivered in two. A hundred and twenty-six million views.

The Palace Museum teamed up with Volcano Engine and "brought back to life" more than 20 artifacts with AI. A Tang dynasty court lady, a bronze vessel — thrown together across dynasties for a Mid-Autumn party. NSR hit 96%.

Good lord. Ninety-six percent.

What that tells you is users aren't rejecting AI content — they're rejecting sloppy AI content.

Story Two: Lenovo Sends the Terracotta Warriors to School

Play Two is even more interesting.

For back-to-school last year, Lenovo put out a short film.

Terracotta Warriors, dressed in traditional robes, wearing backpacks, going to school.

Notice — here AI isn't just an image generator. The fact that the Terracotta Warriors come alive and start talking is itself the core content of this campaign.

Without AI, this film literally couldn't exist.

Tongrentang went even harder. With Kling AI, they built a "pill universe" — tore apart Chinese ancient myth and historical figures and rewrote them into three wildly imaginative little stories. Topic views cleared 2 billion.

Two billion. Let me say that again.

And that's Play Two for AI: let AI be the narrative subject.

It solves a problem that "traditional methods can't tell a new story."

Think about it — what's the biggest pain for heritage brands? Young people can't feel the flavor. A century-old name like Tongrentang lectures about efficacy in product inserts, and young people tune out completely. But turn pills into characters in a universe, bake the efficacy into the plot, and suddenly they'll watch.

Soul and Yebubble took it further still — they ran an "AI boyfriend as offline store manager" pop-up, letting users meet a virtual boyfriend at a milk-tea shop. AI isn't stuck in the screen anymore; it walks right up to you.

Story Three: Mixue's Kuaishou Family Shows Up

Play Three is the most dangerous — and the most tempting.

Last August, Mixue (蜜雪冰城) partnered with Kuaishou and used Kling AI to produce a short film.

But what blew up wasn't the film. It was the user co-creation that came after.

They shipped an "Unexpected Snow King" AI effect. Users type in text, and Snow King transforms.

A thousand users, a thousand Snow Kings.

Topic exposure crossed 1 billion; social interactions hit 771,000.

Over on Douyin Mall, "Call Me AI Edison" went even harder — they put the AI tool directly into the hands of ordinary users. You're not a designer, but you can still turn the idea in your head into a poster. Social interactions: 3.41 million.

And that's Play Three for AI: treat it as a co-creation tool.

It solves a problem of "if I tell the story myself, nobody listens — let users tell it for me."

But.

Every coin has its flip side.

Last year saw plenty of crashes, too.

There was the girl in that airport ad — six fingers on her right hand, five of them fused together. The kind of thing that gives you nightmares.

The Oriental Leaves (东方树叶) outdoor ad, where the model's expression was stiff and her finger joints looked mechanical.

Worse is when the content is just wrong.

A major internet company ran a Super Bowl ad claiming "more than half of global cheese consumption is Gouda." The data was wrong.

A snack brand's product image showed peanuts growing on trees. Peanuts grow on trees.

And some brands stumbled on values. A Chinese tech giant's Qixi Festival (Chinese Valentine's Day) membership copy was flagged as "suggestive" — and the official explanation was that it was AI-generated.

So there it is — the two faces of AI marketing.

It can let By-Health ship a fully AI-generated blockbuster, and it can torch a brand's trust overnight.

What's the Underlying Logic Here?

I thought about this for days, and here's what I landed on.

AI marketing comes in three flavors, but underneath there's only one thing happening: AI is redistributing who gets to have creativity.

It used to belong to the people who could draw, shoot, and write copy.

Now? Creativity belongs to the people who can ask the right questions.

By-Health's team can't paint Antarctica. But they know how to make AI paint the Antarctica they have in their heads. Tongrentang's team can't shoot a sci-fi film. But they know how to make AI turn pills into a universe.

Asking well is worth a hundred times more than executing well.

But.

AI can execute. It can't judge.

It will draw six fingers, because it doesn't know humans only have five. It will say peanuts grow on trees, because it doesn't know botany.

So if you're doing AI marketing in 2026, there are two things you have to hold on to.

One: ask well. The sharper the question, the better the answer AI gives you.

Two: final approval. Before anything ships, a human has to be the gatekeeper — not signing off for form's sake, but actually using their brain.

AI is the productive force — but the responsibility is always human.

One More Thing

Sometimes I think back: five years ago, when we talked about AI marketing, the question was still "can this stuff actually be used?"

By 2025, the question had become "if you don't use AI, can you even survive?"

And in 2026?

I'm guessing the question will become —

"Between you and AI, who's the client?"

I don't have an answer to that one either.

But you'd better start thinking about it now.