---
title: "By 2027, Over 80% of Measurable-Performance Marketing Creative Will Be AI-Generated"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-10-07T14:16:22+00:00"
canonical: "https://ffcap.cn/en/research/by-2027-over-80-of-measurable-performance"
source: "https://uniqueresearch.substack.com/p/by-2027-over-80-of-measurable-performance"
language: "en"
---

# By 2027, Over 80% of Measurable-Performance Marketing Creative Will Be AI-Generated

[![cover](https://substackcdn.com/image/fetch/$s_!5yRM!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F46b32329-fa46-406a-84e9-511273c092f4_2730x1536.jpeg)](https://substackcdn.com/image/fetch/$s_!5yRM!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F46b32329-fa46-406a-84e9-511273c092f4_2730x1536.jpeg)

**35 Minutes a Day: Users No Longer Care Whether It’s AI**

> “When AI can build every dream, authenticity itself becomes the most expensive content.”

Let’s start with two numbers.

The first comes from a roundtable at the Chengdu AI Entertainment Conference: Kuaishou users now spend about **35 minutes a day** consuming AIGC content, and AI-generated work may already account for **30 to 40 percent or more** of newly uploaded content on the platform.

The second comes from another panel. Liu Siyang, founder of Xingbang, judges that by **2027, more than 80 percent of measurable-performance marketing creative will be AI-generated**.

One is the time users spend; the other is the budget clients spend. Both point to the same conclusion: the argument over whether audiences will accept AI content and whether clients will pay for it is already over.

So what is the real question?

The four guests on this panel happen to stand at four positions in the content industry chain: an AI-film-and-TV studio (Yuguang Tongchen / 与光同尘), an agency that spends brand budgets (Xingbang / 星榜), a 3D foundation-model company (VAST), and a state broadcaster with 165 million followers across the web (Sichuan Observer / 四川观察). Together, their remarks draw a fairly complete map of the industry’s next round of competition.

Huang Yuanyuan, partner and COO of Yuguang Tongchen Group, told a dramatic story about the company’s turning point.

In 2023, Yuguang Tongchen was still a traditional TVC advertising house. On a multi-million-yuan marketing project, it was responsible for the art-asset pitch. The project had brought in an art director from Taiwan, but pandemic travel restrictions kept him offline, and the online pitches kept failing.

The reason was a familiar one: traditional art pitches rely on hand sketches and reference images. The client had no picture in their head, so the proposal was castles in the air.

As the deadline loomed, someone on the team used Midjourney to produce a version of the art-asset pitch.

It passed.

Huang Yuanyuan says this left two deep impressions on them.

First, “clients don’t care whether you use AI or some other tool; they care about the final result they see.” Second, AI can land in real commercial projects and produce commercial results. That year-end, the executives met and judged that AI would become the trend across the entire production chain, and decided the whole company would go all-in.

The progress bar that followed is worth noting. In the first half of 2024, combining live-action shooting with AI saved clients about **20 percent** of budget. From mid-to-late 2024, cultural-tourism and product-video projects began delivering on a fully AI path, saving about **50 percent**. By 2025, they were running full-AI pipelines for solutions, including human characters.

Notice the shift: cost saving is only the entry ticket. Huang says the bigger value of AI this year is letting creative ideas actually get monetized — content that used to be impossible because of budget and time constraints can now be made.

“Clients don’t care what tool you use” — every practitioner still debating whether AI content is “low-end” should write that down.

The person most sensitive to where the money flows is Liu Siyang. A post-90s entrepreneur who founded Xingbang in 2016 and has served more than 360 brand clients, she sits right on top of the budget faucet.

Her first observation is counterintuitive: when brands set budgets, they don’t split by “real person versus AI.” They split only by purpose.

Two pockets. One is the **performance budget** with measurable ROI — feed ads, e-commerce displays, Qianchuan (千川) ad buying. The other is the **brand-building budget**.

The performance pocket is being rapidly filled by AI. The logic is blunt: AI creative production is fast, so you can quickly test multiple scenarios; once a video template works, you immediately scale budget behind it.

During the Double 11 shopping festival, livestream rooms need a continuous stream of conversion creative. Shooting can’t keep up, so they lean heavily on AIGC assets.

Consumers buy, so clients stop caring whether the content is AI-generated. That is exactly where the “80 percent by 2027” judgment comes from.

But the other pocket looks very different. Despite plenty of AIGC accounts and viral hits on Douyin and Xiaohongshu, this budget may account for **less than 10 percent** of brands’ total spend — perhaps even lower.

Where’s the gap? Trust.

Liu says when brands pay real creators, they are buying the trust, authentic experience, and professional sharing that a creator brings. The stickiness between fans and a creator comes from trust, and “pure AI currently struggles to replace that.”

In one sentence: AI has already won on “performance,” but on “hearts,” brands are still watching.

Returning to the Kuaishou numbers from the opening — what do they really mean?

Liu Siyang’s read: the supply logic has changed. Models have made content supply extremely abundant. Users scrolling vertically for a few seconds don’t care whether the content was made by AI; only what keeps them matters.

Supply is no longer scarce. Distribution is.

When an audience member asked what the biggest risk is in building an AI-content product today, her answer was traffic and distribution. Compliance, she says, is not the main worry — platform creator rules are already mature and the boundaries are clear.

The real problem: AI live-action dramas, AI comic dramas, and AIGC accounts all erupted this year, and everyone is piling in. The supply of content is enormous. So why should _yours_ be seen?

That is roughly the biggest difference between this wave of AI content entrepreneurship and the last one. The last wave competed on “who can make it.” This wave competes on “who gets selected.”

Will models eventually eat every application? This was the sharpest question of the panel, and the four answers read interestingly side by side.

Li Yingli, head of ecosystem and partnerships at VAST, offered a counterintuitive on-the-ground data point. After GPT-6 Astra launched, many people asked what it would mean for Tripo.

The result: Tripo’s market share wasn’t eaten; its API calls actually grew rapidly — by about **tenfold**, by host Zhao Liang’s account on stage.

Her explanation: stronger general models can execute more complex tasks, which in turn calls more on vertical-model capabilities. 3D foundation models won’t be fully replaced by large language models.

What models eat is low-value, repetitive tools and workflows — not vertical capability itself.

Liu Siyang agrees on the broad “eating” direction. She estimates 80 percent of generic input-output work will be absorbed, and AI video production will keep getting simpler. But she isn’t anxious: “aesthetic judgment and experience are hard to replace.”

AI can do data analysis and generate video, but it struggles to tell you how to set an AIGC account’s persona so users like it, or what story will resonate. “It can execute; choosing the right thing to do still needs human judgment.”

Huang Yuanyuan added an organizational observation: Yuguang Tongchen’s talent needs keep shifting.

First they wanted AI image generators; after Seedance 2.0, the workflow changed and execution directors with a director’s thinking became more valuable. People who only generate images and write prompts must move toward being an AI supervisor who coordinates a whole project.

What AI replaces is the production step; what remains is judgment. Her words are worth quoting directly:

“What gets replaced may be the batch, industrialized parts of production; good stories, good content, and creators and directors with ideas will become more and more precious.”

Liu Qing, head of Sichuan Observer’s AIGC division, took the question from a completely different direction: authenticity.

AI can already pass as real. Once that happens, where do people go to find real content, and how do they tell true from false? Fictional content is “dream-making” — long dramas, short dramas, and films will gradually be rewritten by AI. But real content, precisely because of AI, becomes scarcer. And that is exactly what mainstream media holds.

A vivid detail: Sichuan Observer built a virtual artist, Una Yiyi (伊一), and held an online concert for her at the end of last year.

The station’s human hosts were resistant at first, then found that their own images could appear, through the virtual artist, in big productions they’d never had the budget for. Gradually they accepted it.

AI makes things possible that used to be impossible — and traditional media may feel that no less than film studios do.

Put the four voices together, and the panel’s through-line is clear.

When content can be produced infinitely, production itself becomes less valuable. Value runs to two ends: on one end, taste, judgment, and good stories; on the other, distribution, trust, and authenticity.

Yuguang Tongchen’s slogan is “let good stories shine.” Huang Yuanyuan says the point is still the good story — it’s just that the “shining” is becoming easier and more diverse.

And Liu Qing leaves us with the line that can serve as this piece’s footnote: when AI can build every dream, authenticity itself is the most expensive content.

* * *

-   **Huang Yuanyuan**, Partner & COO, Yuguang Tongchen (与光同尘)
    
-   **Liu Siyang**, Founder & CEO, Xingbang (星榜)
    
-   **Li Yingli**, Ecosystem & Partnerships Lead, VAST
    
-   **Liu Qing**, Head of AIGC, Sichuan Observer (四川观察)
    

**Zhao Liang**, Partner, Extraordinary Capital (非凡资本)

**Zhao Liang:** Over the past year, I’ve talked at many AI events about “AI rewriting the content industry.” AI content is also the area where commercialization is most visible right now. We held our AI Entertainment Conference in Chengdu and need to study this direction closely. Let’s have the four guests introduce their businesses and their niche in AI content. Ms. Huang first.

**Huang Yuanyuan:** I’m Huang Yuanyuan from Yuguang Tongchen. We are an AI-native film-and-TV company. On top of our in-house Yuguang AI and our production capability, we connect the commercial chain from content production and film/TV distribution to IP building. Our business spans AI film, AI animated comic-dramas, AI commercial advertising, AI software training, and overseas business.

**Zhao Liang:** Yuguang is a typical AI content production company. Next, Ms. Liu.

**Liu Siyang:** I’m Liu Siyang, founder of Xingbang. We’re a data-driven, full-funnel marketing service provider. Core business includes brand content marketing, e-commerce operations, and our Chengdu MCN “Xingbang Huyu.” In AIGC, we help clients produce and place AIGC creative.

**Zhao Liang:** Ms. Liu knows exactly where client budgets flow. Next, Sister Ka (Li Yingli).

**Li Yingli:** I’m from VAST. We’re an AI company building general foundation models; early on we focused on AI 3D, and this year we also released a world model. On top of our self-developed AI 3D and world models, we’re building interactive creation for the general public to lower the barrier to creation. We have two product directions: Project Eden, a world-model research project released this year, and Tripo Studio, our one-stop AI-driven 3D workstation where users can go from generation, part decomposition, topology, and binding through to application.

**Zhao Liang:** Next, Mr. Liu from Sichuan Observer.

**Liu Qing:** I’m Liu Qing, head of AIGC at Sichuan Observer. We’ve been creating AIGC content since 2024 and want to exchange experience today.

**Zhao Liang:** This event also partnered with Sichuan Observer. I just checked the back-end: views of the Chengdu AI Entertainment Conference have already passed 70,000. Ms. Huang, what made you transition from a traditional ad company to AI content, and how did you find today’s model?

**Huang Yuanyuan:** As the host said, we were a TVC production company — this is our 11th year, moving from post-production to production to full-funnel marketing. In 2023, on a multi-million-yuan traditional project, we were responsible for the art-asset pitch. A Taiwanese art director couldn’t come offline due to the pandemic, and the online pitch kept failing. Traditional art pitches use hand drawings and references; they can’t make the concept concrete the way AI can today. With the deadline near, a team member used Midjourney to make an art-asset pitch — and it passed. That was also the year Midjourney and Stable Diffusion exploded.

It gave us two impressions: clients care about the result, not the tool; and AI can land in commercial projects and deliver results. That year-end, the executives decided to go all-in. But in early 2024, AI was far weaker. In the first half we still mixed live-action with AI, saving clients about 20 percent. From mid-to-late 2024, some cultural-tourism and product-video projects delivered fully on AI, saving about 50 percent. By 2025, including human characters, we used full-AI pipelines. This year the value is not just saving budget; it’s letting creators monetize ideas, widening content boundaries and reducing homogeneity.

**Zhao Liang:** Sister Ka, do you define yourself as a model company, or a product/app/platform company?

**Li Yingli:** We’re not just a model company, not just an app or platform. We’d say VAST uses self-developed AI 3D and world models as the base, so that AI-generated 3D content actually enters production workflows. We started with 3D generation, then built the Tripo Studio workstation, then added plugins, and the ecosystem came together.

**Zhao Liang:** Tripo now has tens of millions of creators and among the world leaders in 3D model generation volume. From an ecosystem standpoint, which industry’s demand surprised you most?

**Li Yingli:** Over the past year gaming was a clear direction, tied to generation capability. But as models iterated, we saw e-commerce, 3D printing, spatial intelligence, industrial, and offline large-space XR demand growing fast too.

**Zhao Liang:** By 2027, what’s the next breakout scene — gaming, e-commerce, or short video?

**Li Yingli:** We’re more bullish on production-grade 3D-asset workflows and the interactive-world production that world models enable. 3D-native interaction has lots of headroom; e-commerce, 3D printing, XR, and film production benefit accordingly. For example, we have high-poly and low-poly products. Low-poly was first shown at GDC this March; high-poly assets can enter industrial-grade applications, even 2A-plus games, and 3D printing. Low-poly used to have topology problems; from Tripo P1.0 in March to P2.0 last month, that’s largely solved, and generated assets can enter mobile games, VR, and offline large-space XR.

**Zhao Liang:** Ms. Liu, you founded Xingbang in 2016, have a big-tech background, and have served 360-plus brand clients. As AI content grows, how do brands view AI versus real-person content, and how will they allocate budget?

**Liu Siyang:** We’ve felt this deeply. But when brands set marketing budgets, they don’t start by splitting real versus AI; they start from their own purpose. There are two budget types: performance budget with measurable ROI, and brand-building budget. From 2025 until now, in measurable performance budget, feed ads, e-commerce displays, and Qianchuan creative have already been heavily replaced by AI. I think by 2027, over 80 percent may be AI-generated. Clients found consumers will pay, so they stop caring whether it’s AI; AI creative is faster and lets you test scenarios quickly, and once a video template works you scale budget fast. Especially Double 11, when livestream rooms need a steady stream of conversion creative and shooting can’t keep up — they use lots of AIGC. The market education moved fast.

But in brand-building budget, even though there are already many AIGC accounts and hits on Douyin and Xiaohongshu, this is still a very small share of brand clients’ total budget — maybe under 10 percent, or lower. We’re bullish on the direction, but brands are more rational now. Choosing an AIGC account isn’t just about the novelty of AI; it’s about reach, whether it conveys brand value and influences consumers. AI accounts with strong data and IP character will have more room.

**Zhao Liang:** Follow-up: this morning a Kuaishou guest shared two shocking numbers — users spend about 35 minutes a day consuming AIGC content, and 30 to 40 percent or more of new content is AI-generated. What does that mean?

**Liu Siyang:** First, the supply logic changed. Model development made supply very abundant; in the end it’s about what really creates stickiness — maybe fast-paced AI drama, maybe crafted design. At its core, can the content attract users. Especially in a few vertical seconds, users don’t care much whether it’s AI; only retaining them has value.

**Zhao Liang:** Mr. Liu, Sichuan Observer is a mainstream outlet with 165 million followers online, yet not the rigid official media people imagine. How do you position yourselves and build the AIGC business?

**Liu Qing:** The first answer is simply “speak human.” We convert serious content into forms people enjoy watching and listening to — that’s how we attract users. Second, we’re inclusive toward models and new technology. Although we’re in the broadcasting system, we don’t reject new things. We have dedicated people testing newly released models and technologies as technical reserve, then combine them with our existing production capability to optimize the workflow. We’ve done AIGC since 2024 — commercial films, ads, virtual artists. We found some external tools not handy, so we built our own internal platform, and are now considering opening it to others.

**Zhao Liang:** Tell us about your virtual artist Una Yiyi and her online concert.

**Liu Qing:** Our station already has many hosts; adding a virtual artist wasn’t easy to understand at first. Years ago there were “digital humans” — record a host for ten minutes, vendor generates content the next day — but those looked too fake. When we started doing AI content, we found virtual artists were feasible. Una Yiyi did an online concert at the end of last year and it went well. More importantly, she can be a bridge connecting the station’s human hosts with virtual content. Hosts were a bit resistant at first, then found their own images could make content with the virtual artist that they couldn’t afford before — big productions, dramas, TV shows, ads, even films. Hosts already need to promote themselves, and now they have more forms, so they’re gradually accepting it. The virtual artist isn’t just herself; she drives virtual-real combination. We’ve also tried using Tripo to make 3D models and 3D-print them, hoping the virtual and real worlds connect. VR may not be dominant right now, but we remain bullish on lighter tools that let a virtual artist and real people be sensed more truly in the same space.

**Zhao Liang:** The same question to all four: as models get stronger and iterate faster, will they eventually eat every application? You sit at content production, client services, the model layer, and mainstream media respectively — how do you stay competitive? Ms. Liu first.

**Liu Siyang:** On “models eating everything,” we also think 80 percent of generic input-output work will be absorbed. AI video production will get simpler, and the canvas may not be the final form. But content production has a place to play. Aesthetic taste and experience are hard to replace. AI can do front-end data analysis and generate video, but it struggles to tell you how to set an AIGC account’s persona so users like it, or what story resonates. It can execute; choosing the right thing still needs human judgment. To stay competitive, technology moves so fast we can only move carefully. The core is deepening understanding of clients and consumers, and using AI tools plus human experience to turn vague client needs into deliverables. From a brand-marketing-services view, content production is only one step; we’re also building our own Agent connecting front-end insight, content production, and投放 feedback, turning each real project’s experience into Skills and templates, then into systematic product engineering.

**Zhao Liang:** So you think models absorb lots of generic work but you’re not that anxious about your own business?

**Liu Siyang:** Right. It’s ultimately general intelligence, but personalized, differentiated content production is itself a personalized, differentiated process; it can’t fully replace it. We’re not too anxious.

**Zhao Liang:** Ms. Huang?

**Huang Yuanyuan:** From a bigger view. Last week I attended the Chongli Forum, where there was an AI Builders section with lots of Gen-Z creators and ideas about physical AI. AI will become more and more pervasive and invisible in life, work, and business — not as obvious as wearing a VR headset to watch video today, but maybe built into ordinary glasses showing digital information, or a small alarm clock monitoring sleep. It’s irreversible and brings benefits. Right now we still do some “beast-of-burden” work; in the future that may not be needed, especially for young people, who can focus on how to use AI and benefit from it. Back to our industry, our org keeps adjusting: the pipeline first needed AI image generators; after Seedance 2.0, execution directors with a director’s thinking became more suited to commercial film projects. People who only generate images and write prompts must move toward AI supervision and coordinating whole AI projects. Future talent needs to be more compound. AI also replaces some repetitive labor — for example, a character’s art assets still need constant generation, review, and storage, which will gradually be replaced. I think that’s good.

**Zhao Liang:** Sister Ka? You build models yourselves.

**Li Yingli:** I have to be careful here. We believe models won’t simply eat everything; they’ll eat low-value, repetitive tools and workflows. As model capability iterates, competition goes deeper. For example, after GPT-6 Astra came out, many asked its impact on Tripo. It didn’t eat Tripo’s share; on the contrary it had a positive effect, and our API calls multiplied. So 3D foundation models won’t be fully replaced by LLMs; LLMs let our model enter more production workflows. As Astra executes more complex programs, Tripo’s capability can be used more deeply. If our 3D structure generation, editing, and repair keep improving, trial-and-error and repair cost drop, which promotes co-development of products like Tripo and Astra.

**Zhao Liang:** A money question. VAST uses SaaS subscription overseas and some free strategies domestically. Where will model-layer money come from — subscription, API, or content itself?

**Li Yingli:** Future model-layer revenue won’t come only from SaaS or only from API. SaaS gives creators continuous tools; API gives developers and enterprises creation capability to cut cost and lift efficiency. We don’t just sell a generated 3D model. Combining self-developed AI 3D with world models, the goal is an interactive generative-content platform. Models can enter e-commerce and 3D printing first, but applications are just the first step. Later we’ll connect more production workflows so generated content is used more smoothly; these 3D assets can be imported into world models so content is generable, interactive, controllable, and continuously operable. We want an aggregated, interactive 3D content creation platform.

**Zhao Liang:** Mr. Liu, same question from mainstream media’s view.

**Liu Qing:** Short term versus long term. In the short term it can’t replace, but it will gradually optimize workflows. Professional production can use AI to merge three roles into one, three steps into one. Another category is the market growing wild first and then regulating. Listening to everyone, I thought of a long-unmentioned feature: authenticity. AI already passes as real; after something happens, where do people find real content and tell true from false? Without media like us, it gets harder. So at least in the near term, media still has a role. On the other side, fictional content is “dream-making” — long dramas, short dramas, films — and I feel much of that will be taken over by AI.

**Zhao Liang:** Last question: by 2027, what part of the content industry — step, job, or business model — gets fully rewritten by AI first? Ms. Liu will take the audience question afterward. Ms. Huang first.

**Huang Yuanyuan:** I hope people don’t over-anxiety. Steps or jobs being replaced, or business models changing, are all improvements and iterations on top of existing industry know-how. The key is how AI empowers the content industry. The first wave to move into AI were often already excellent directors, cinematographers, or post specialists; AI removed the technical wall. Before, a project might need another team; now they can realize their own ideas. Yuguang’s slogan is “let good stories shine” — the point is still the good story, and the shining can be easier and more diverse. Good content will get more diverse; not only content with good commercial results counts. Good aesthetics, good style also count; art creation needn’t be measured only by commercial results and is equally precious. What gets replaced may be batch, industrialized production; good stories, good content, and creators and directors with ideas will become more precious.

**Zhao Liang:** Sister Ka?

**Li Yingli:** What AI rewrites first won’t be creative jobs, but repetitive labor. In content-industry terms, it’s the production step between an idea and a usable asset. In 3D, it used to take many people collaborating on modeling, topology, texturing, editing, rigging, and smart decomposition. After AI, the workflow is redefined: from text or image, directly generate an editable, usable 3D model that enters the downstream workflow.

**Zhao Liang:** Mr. Liu?

**Liu Qing:** Everyone has weaknesses, and AI can help fill them. But in practice, one person running all production isn’t realistic. I think the future may be a few strong-capability people working together. In making films we feel strongly: how you narrate and whether you have aesthetics increasingly determine content quality. Using AI to raise that capability is the direction.

**Zhao Liang:** Ms. Liu, please answer the audience questions — which one do you choose?

**Liu Siyang:** The first two are more relevant to me, so I’ll take those. The biggest risk in building AI content products now is traffic and distribution. Content supply is just too abundant — this year AI live-action dramas, AI comic dramas, and AIGC accounts all erupted at once. Raising distribution efficiency, and which aesthetics and stories attract users, is the key competition. As for compliance, platform creator rules are already fairly mature; as long as you follow the rules, the boundaries are clear, so I’m not too worried.

The second question we often discuss with clients: clients aren’t yet investing heavily in AIGC KOLs, an important reason being that they want the trust a creator brings — authentic experience and professional sharing. The stickiness between fans and creators comes from trust; pure AI currently struggles to replace that. As for lots of “pickled mustard” content, users’ own judgment and platform algorithms balance it. Users have taste and judgment; good content and content with real experience will ultimately be seen more. AI helps us make content that was hard and expensive to shoot live, but authentic-feeling content will keep existing.

**Zhao Liang:** Ms. Liu mentioned that after AI rewrites the content industry, traffic and distribution efficiency become a key risk — which leads into tomorrow’s discussion of overseas expansion and globalization: which market and platform you enter first is all tied to distribution efficiency.

[![cover](https://substackcdn.com/image/fetch/$s_!5yRM!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F46b32329-fa46-406a-84e9-511273c092f4_2730x1536.jpeg)](https://substackcdn.com/image/fetch/$s_!5yRM!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F46b32329-fa46-406a-84e9-511273c092f4_2730x1536.jpeg)

---

Original publication: https://uniqueresearch.substack.com/p/by-2027-over-80-of-measurable-performance
On-site reading page: https://ffcap.cn/en/research/by-2027-over-80-of-measurable-performance