The most important AI marketing story in China is not that Chinese models are becoming smarter. It is that the distance between a consumer’s sentence and a completed order is collapsing.

A shopper describes a situation. An AI assistant clarifies the need, compares products, identifies a promotion and places the order. There may be no keyword search, no advertising click, no product-page session and no conventional shopping cart.

When that journey compresses into a single conversation, the basic unit of digital competition changes. It is no longer simply the keyword, the impression or the piece of content. It becomes the machine’s judgment about which product best satisfies an expressed intention.

For Western executives, China’s significance lies less in the individual platforms involved than in what they reveal about the next operating model of commerce: marketing is moving from influencing human attention to shaping machine-mediated decisions.

The recent catalyst is not another chatbot launch

On 15 October, Alibaba placed AI-assisted shopping at the centre of this year’s 11.11 festival. Tmall said the event would bring together one million global brands and more than 70 million products, with the consumer experience built around the integration of Taobao and Qwen AI.

That does not mean one million brands suddenly have an AI strategy. It means AI-mediated discovery is being introduced on a commercial stage where global brands already compete, launch products and allocate substantial promotional budgets.

The underlying integration is extensive. In May, Alibaba connected Qwen to Taobao and Tmall’s full catalogue of more than four billion products. Users can browse, compare, place orders and manage deliveries conversationally, while agents support order management, logistics and after-sales service.

The strategic asset here is not Qwen alone. It is Qwen’s proximity to catalogues, reviews, transaction histories, payments, fulfilment and merchant operations. A model can generate a recommendation. An integrated commerce system can determine whether that recommendation is available, affordable, deliverable and worth trusting.

That difference matters.

China’s platforms are racing to own the intent-to-transaction layer

Alibaba is not an isolated case.

Ant Group said Alipay AI Pay had surpassed 100 million users by 23 February and processed more than 120 million transactions during the week of 5–11 February. Its payment product allows consumers to transact through AI agents without switching between pages.

ByteDance has also been testing integration between Doubao and Douyin e-commerce, allowing users to place orders and complete payment without leaving Doubao.

Tencent’s Yuanbao has taken a different route through JD.com. A shopping query can produce a JD product card inside the AI conversation, although product selection and checkout still require user action in JD’s mini-program.

These implementations are not equally mature. That is precisely why executives should pay attention. The market is showing the intermediate stages through which consumer behaviour is likely to develop: from answer, to recommendation, to product card, to supervised checkout and, only in some scenarios, to delegated purchasing.

The merchant side is changing at the same time. Tmall is adding agentic capabilities to its Business Advisor tool, covering store analytics, advertising placement, customer service and post-sale processes. Alibaba says its Dianxiaomi customer-service AI had been piloted with 200,000 merchants, producing a reported 30% conversion lift.

A consumer agent on one side. A merchant agent on the other. Between them, the traditional marketing funnel starts to look less like a funnel and more like a negotiation between software systems.

Why China is moving faster

China’s advantage is structural, not mystical.

Several of its largest digital groups can connect an AI assistant to product catalogues, payments, order histories, customer reviews, merchant tools and fulfilment without assembling a new coalition for every transaction. They can observe not only whether a recommendation generated a click, but whether it produced an order, a return, a complaint or a repeat purchase.

That creates a tighter learning loop.

Western commerce is more fragmented across independent retailers, payment providers, marketplaces, advertising systems and brand websites. This fragmentation has benefits—consumer choice, merchant autonomy and less dependence on a single platform—but it makes end-to-end execution harder to coordinate.

Chinese consumers also already live with algorithmic feeds, social commerce, mini-programs and integrated mobile payments. Delegating another step to software is a smaller behavioural leap than it would be in an environment built around separate websites and manual checkout.

There is a commercial incentive as well. Platforms need new sources of growth. Merchants want lower operating costs and better conversion. AI offers both sides something valuable: the platform gains a more powerful gateway to demand, while the merchant receives automation across marketing and operations.

The agent is therefore not merely a new user interface. It is an instrument of platform power.

The West is building a different architecture

It would be a mistake to present this as China advancing while the West stands still.

OpenAI has introduced Instant Checkout through its Agentic Commerce Protocol, beginning with US Etsy sellers and announcing Shopify merchants including Glossier, SKIMS, Spanx and Vuori as future participants. Merchants continue to handle payment, fulfilment and customer support.

Google’s Universal Commerce Protocol connects businesses with AI agents across the shopping journey. UCP-powered checkout is rolling out for Etsy and Wayfair purchases inside AI Mode in Search and the Gemini app, with Shopify, Target and Walmart announced as subsequent participants.

Microsoft is rolling out Copilot Checkout through PayPal, Shopify and Stripe, with Urban Outfitters, Anthropologie, Ashley Furniture and Etsy sellers among its launch merchants. Its design explicitly keeps the retailer as merchant of record and preserves ownership of the transaction and customer relationship.

Two architectures are emerging.

China is demonstrating the vertically integrated agent: one ecosystem can connect intent, recommendation, payment and service. The West is placing greater emphasis on protocols that allow independent agents, merchants and payment systems to interact.

The first model may learn faster inside a closed environment. The second may offer greater interoperability and merchant control. Neither is universally superior.

For a global CEO, the lesson is not to copy Alibaba’s channel plan. It is to recognise that the future of digital marketing will depend as much on transaction architecture as on creative content or media buying.

Five assumptions Western companies need to abandon

1. “Our website is where conversion happens”

Increasingly, the decision may be made before a customer reaches the website—or without a website visit at all.

The AI-generated shortlist becomes a new kind of shelf. If a brand is absent from the three options presented, the quality of its product page is irrelevant. If its price, availability or warranty information is wrong inside the conversation, the customer may never reach the brand’s owned channel.

This does not make websites obsolete. They remain important sources of verification, product truth and service. But their role in the journey is changing.

2. “GEO is primarily a content problem”

Generative engine optimization can help a brand become visible in AI answers. Agentic commerce raises a harder requirement: can the brand be selected and transacted?

An agent needs more than polished articles. It needs intelligible attributes, compatible variants, current inventory, delivery terms, return rules, warranty coverage, certifications and credible evidence behind product claims.

In that environment, the product record becomes a form of media. A mistranslated specification or inconsistent return policy is no longer back-office hygiene. It can become a visibility and conversion problem.

3. “Marketing and operations can remain separate”

An AI agent does not experience a campaign in isolation from the commercial reality behind it. Price, stock, delivery speed, reviews and service conditions can all influence whether a recommendation becomes an order.

That erodes the old division in which marketing generates demand and operations deals with everything after the click. The CMO’s performance increasingly depends on data and processes owned by e-commerce, IT, supply chain, finance and customer service.

AI readiness is therefore an organisational issue before it is a media issue.

4. “AI performance can be measured through referral traffic”

Adobe reported that US retail traffic from AI sources grew 62% year on year in July, while AI-referred visitors converted at a rate 60% higher than non-AI traffic.

Those figures demonstrate the value of AI-mediated discovery. But referral metrics will miss transactions completed entirely inside an AI interface.

Companies will need a wider measurement system: inclusion in relevant prompts, position within the consideration set, factual accuracy, agent-assisted sales, contribution margin, return rates and the quality of the resulting customer relationship.

Last-click attribution cannot explain a purchase that never produced a conventional click.

5. “Consumers will either use AI shopping or reject it”

The more important distinction is between influence and autonomy.

NIQ’s consumer research found that around 34% of consumers used AI to research products, but only about 8% had allowed AI to complete a purchase for them.

Autonomous shopping is not about to replace human judgment across every category. Consumers are far more likely to delegate comparison, review summarisation, coupon selection and routine replenishment before handing over high-value or emotionally significant decisions.

Companies that over-rotate toward fully autonomous commerce will misread the near term. Companies that ignore AI’s growing influence over consideration will misread the structural shift.

What CEOs should do in the next 90 days

This does not require a three-year transformation programme. It requires a disciplined test of whether the business can be understood, recommended and served in an AI-mediated environment.

Start with the machine’s view of the category. Build a set of 50 realistic Chinese-language prompts reflecting customer situations rather than brand keywords. Run them across the AI services relevant to the business, including Qwen, Doubao, Yuanbao and other domestic discovery surfaces. Record which competitors appear, which sources are cited, what claims are made and where a transaction can occur.

Create a controlled product-truth layer for a priority category. Select a manageable group of strategically important products. Reconcile their Chinese specifications, use cases, compatibility, inventory, pricing, delivery, returns, warranties and certifications across marketplaces, distributors and owned channels. Legal and regulatory teams should approve the claims before marketing expands their distribution.

Run one closed-loop pilot. Choose a category where the team can connect AI visibility with commercial outcomes. Establish a baseline, correct the product information, improve supporting evidence and test the result. Measure margin and returns alongside recommendation frequency and sales. Publishing more content without changing the underlying product data is unlikely to be enough.

Assign an accountable owner. AI commerce cannot sit indefinitely between the China e-commerce team, global marketing and IT. One senior owner should coordinate product data, platform relationships, measurement, legal review and customer service. The budget can come from existing search and retail-media experimentation; this should not begin as another isolated innovation project.

Use China as a global learning system. The lessons from Chinese closed ecosystems should inform preparations for Google UCP, OpenAI’s commerce protocol, Copilot Checkout and future merchant agents in Western markets. Product-feed quality, machine-readable policies, merchant-of-record decisions and customer-data rights all transfer across regions.

What happens next

Over the next 12–18 months, the most valuable deployments will probably remain supervised rather than autonomous. AI will compare products, interpret specifications, summarise reviews, assemble offers and guide checkout. Routine replenishment and service transactions will move faster than luxury, healthcare and other high-consideration categories.

Platforms will compete on the quality of their commercial context, not just model benchmarks. Expect more sponsored offers and transaction-based monetisation, alongside pressure for clear disclosure when commercial incentives influence a recommendation.

Brands with accurate data, dependable fulfilment and strong official stores should have an advantage. Those relying on fragmented distributors and inconsistent product information will find that AI exposes weaknesses previously hidden by search rankings and campaign creative.

Over three to five years, consumer agents and merchant agents are likely to negotiate more of the transaction: bundles, loyalty benefits, delivery options, replenishment timing and personalised offers. China’s integrated ecosystems and the West’s protocol-based networks will continue to coexist, each borrowing ideas from the other.

The resulting competitive divide will not simply be between companies that “use AI” and those that do not. It will be between companies whose products can be confidently evaluated by machines and companies whose value remains difficult to express outside human-controlled channels.

The boardroom question is therefore larger than whether to add GEO to next year’s marketing plan.

Can your company be understood, trusted and purchased through an AI agent—without surrendering the customer relationship to the platform that controls the conversation?

China offers an early view of that challenge. Western executives should treat it as a strategic rehearsal, not a regional curiosity.