Taobao Wants Every Product to Become Its Own Content

E-commerce advertising has a shape everyone in the industry knows. You buy an impression, a user sees your product, and the money is gone whether or not anything happens. The ad performs or it does not, and next month you buy more.

Taobao laid out a different idea at a closed-door AIGC ecology meeting in Hangzhou on 22 September. The phrase its content chief, Dai Yan, used was "product-native content, one product many narratives." The claim underneath is that the product stays true while the story around it changes to fit the person watching. In Taobao's framing: expression varies, the product does not.

One product, many narratives

Taobao describes its content model as moving from "a thousand faces" to "one product many narratives." The older phrase described the recommendation system deciding what to show a given user. The new one describes the content itself adapting. The product is a fixed point. The scene, the characters, and the emotional angle are variables.

That distinction matters because it is the thing generative AI is genuinely good at. A brand with one product and twelve audience segments needs twelve versions of the same message, and until recently producing twelve versions cost twelve times as much as producing one. If the product's factual content has to stay identical across all twelve, that is a constraint a model can respect. The narrative around it is the part that changes.

Taobao's own summary of the principle was blunt about the risk it is trying to avoid. The product has to stay consistent, because the moment an AI-generated scene changes how a product looks or works, the content has stopped being advertising and started being fiction. Dai Yan framed the whole push as starting from the product, keeping truth at the centre, and landing on the transaction. That last part is what separates this from a content play. It is not trying to be entertaining. It is trying to convert.

The Sweet Orange agent

The tooling behind the push is a content agent called Sweet Orange, launched on Taobao's Guanghe creator platform after 24 September, with apparel merchants as the first group. It covers the whole chain: picking products, generating creative, producing assets, distributing them, and measuring what happened. Taobao is also adding AI diagnostics and cleaner reporting, so a merchant can see which variants worked instead of guessing.

The distribution side is where the model gets unusual. Taobao is promoting a pattern it calls short drama, short video, live stream, treating the three as stages of one piece of content rather than three separate campaigns. A short drama sets up a story. Short videos cut it into pieces. The live stream sells against it. The same material feeds all three, which is the economic argument: produce once, reuse across formats.

Taobao also stood up an alliance for e-commerce short drama with partners including iQiyi, CNR, Migu, and Fengmang, and attached a fund it describes as a hundred-billion-yuan class of support resources. Merchants can join through a product-linked model where product subsidies, red packets, and channel-exclusive pricing serve as their entry stake. The pitch is a merchant spends once and gets distribution across two platforms.

The roles Taobao is trying to create

The plan reaches past tools. Taobao says it wants to incubate a set of new roles around e-commerce content: short drama service providers, UGC creators, AIGC service providers, and digital human vendors. Read that as an admission that the platform cannot supply all the content itself. If merchants are going to treat content as an asset, someone has to make the asset, and Taobao is betting a services layer forms around the tools rather than the tools replacing it.

That has a familiar precedent in Chinese e-commerce. When Taobao opened its ecosystem to third-party service providers years ago, a large industry of shops grew up around it, each selling a narrow service to sellers who could not do the work in house. The AIGC push is an attempt to repeat that at the content layer. Whether it works depends on whether AI-generated drama is a service a merchant will pay a specialist for, or a button a merchant presses alone.

The bubble this sits on top of

Here is the part the cheerful numbers leave out. China's micro-drama industry put out roughly 367,000 titles in the first half of 2026. About 75 percent were AI-made. Fewer than 0.5 percent of titles crossed a hundred million plays. Traffic costs rose more than 100 percent year on year while revenue fell by more than half.

Read those four numbers together and they say something simple: AI collapsed the cost of making content and did nothing to the cost of getting it seen. Production was never the bottleneck. Attention is. A market where three quarters of the output is machine-made and almost none of it reaches an audience is not a market where making more content helps.

Taobao's answer is to stop treating content as a one-off campaign and start treating it as an asset. A short drama that a merchant owns can be re-cut, re-scored, and re-published for months. An ad impression is consumed the moment it is bought. In that framing the value is not in any single piece of content. It is in whether the merchant accumulates something that keeps working.

Whether that holds up is an open question. An asset only appreciates if people watch it, and the same saturation that crushed the standalone micro-drama market applies here. Taobao has an advantage the micro-drama studios do not: its content sits next to a buy button, so a modest audience can still convert. That is a real structural difference, and it is also the reason the company keeps saying "landing on the transaction" instead of talking about reach.

What this says about AIGC in commerce

The larger signal is that e-commerce is becoming the least speculative market for generative video. Advertising has a measurable return, so a merchant can look at a hundred AI-generated videos and decide whether the spend paid for itself. Compare that with a short drama studio hoping for a hit, where the feedback loop is slower and the outcome binary. When the return is calculable, adoption does not need to be argued for.

It also puts pressure on the tools. If a merchant is going to treat content as an asset, the asset has to stay consistent with the product across every version. That is a harder technical problem than making a good-looking video, because it means the model has to know what about the product cannot change. A scene can be reimagined freely. A bottle's shape cannot.

Taobao's bet is that the constraint is a feature. By fixing the product and varying everything around it, the company gets the volume that AI makes cheap and the consistency that commerce demands. It is a narrow path between a factory that produces nothing anyone watches and a machine that generates pretty scenes that misrepresent what is for sale. Whether a hundred billion yuan of support resources can widen that path is what the next few quarters will show.