Assistants reroute the marketplace ladder
2026-11-12BXT Data Editorial

Assistants reroute the marketplace ladder

Assistants reroute the marketplace ladder article image

Sunny He, a 32-year-old shopper in Shanghai, made one purchase this Singles Day, a 12,000 yuan iPhone 18 Pro Max with a 300 yuan voucher, and said the festival no longer feels lively enough to browse. Her caution is the demand side of a platform problem: coupon math was deleted, assistants were pushed into the funnel, and the 1.02 trillion yuan booked across major platforms by October 31 says less about appetite than about which gateway produced the order, up to November 12, 2026.

The catalogue is no longer the shop window

Known factsUnverifiedNeeds verification
Cumulative sales across major platforms reached 1.02 trillion yuan in the three weeks to October 31Whether the total includes returns or cancellationsThe full-period figure released after November 12
Roughly 5 percent of Douyin purchases in October came through chatbot links, over 20 percent for supplementsWhether the share is measured on orders or on valueRepeat purchase rate inside assistant traffic
JD's shopping assistant drew nearly 80 million users, more than three times last yearHow many of those users completed an orderBasket size against non-assisted sessions
Tmall reported AI token sales up more than 700 percent between October 15 and November 5Whether the base period was promotionalThe renewal rate on monthly token subscriptions

The structural numbers sit behind that single purchase. Total retail sales rose 1.1 percent in the first nine months of 2026, online sales of goods rose 3.8 percent and online sales of services 5.3 percent, according to the National Bureau of Statistics. Against that, the 2.4 percent growth in festival sales to October 31 is modest, and it arrived after platforms removed the layered discount structures that used to generate the concentrated surge.

What replaced the coupon math is a prompt. Alibaba put Qwen tools inside Taobao and Tmall for research, virtual try-on and discount discovery, and JD's assistant handled nearly 80 million users looking for advice. On Douyin, about 5 percent of October purchases came through chatbot links, and in supplements the share passed 20 percent. Those are small numbers with an unusual property: they are decided before a shopper ever sees a listing page.

GatewayThis season's numberWhat a brand can seeWhat it cannot see
Assistant chat inside the appAbout 5 percent of Douyin purchases in OctoberThat an order was assistant-ledWhich prompt produced the shortlist
Considered categoriesOver 20 percent for supplementsCategory-level swing toward assistantsWhether the brand was shortlisted or substituted
Platform assistant at scaleNearly 80 million JD assistant usersReach of the entry pointConversion by category inside it
Subscription AI servicesToken sales up more than 700 percentThat a new product line existsWhether buyers renew after the first month
Cross-border listingAliExpress warehouses in six countries, pre-event stock up 146 percentInventory positioning by marketSell-through per market

Assistant gateways already carry volume

Five percent on Douyin, twenty on supplements

The gap between 5 percent and 20 percent is the interesting part. In categories where the buyer needs help comparing, an assistant is a natural intermediary, and the brand that is not in the shortlist does not lose a ranking, it loses the consideration set. In categories where the buyer already knows what they want, the assistant is a checkout shortcut and the shelf still decides. Category structure, not platform policy, is what sets the exposure.

This is a consumer demand story with a data layer on top. Demand did not collapse; it moved to a gateway that compresses research into one exchange and then buys. Every shop that spent the last decade optimising a listing page for a human scrolling a grid now has to be legible to something that reads attributes, compares three candidates and never scrolls. Category growth in the next cycle will partly be decided by that legibility.

Tokens became a product line

One hundred thousand subscriptions in half an hour

The strangest result of the season is that AI became inventory. Tmall reported token sales up more than 700 percent between October 15 and November 5, AI toys passed 100 million yuan within a year of becoming a category, and companion robots rose nearly fivefold in the second half of October. A livestream sold 100,000 monthly subscriptions to Alibaba's QwenWork in half an hour at a promotional price of 0.01 yuan each.

A subscription sold at one fen is a customer acquisition cost, not a price, and that is the frame worth keeping. The season established that Chinese shoppers will pay for AI services in small amounts and in bundles, which is a different question from whether they will renew. For a marketplace, tokens and subscriptions are attractive because they are repeatable and measurable; for a brand, they are a reminder that the same assistant recommending a product can also be the product being sold.

  1. Log every session that arrives from an assistant with a referral parameter, and keep it as its own segment
  2. Record whether the brand was named, shortlisted or linked, as three separate events
  3. Compare basket value and return rate between assistant-led and browse-led orders
  4. Freeze the product attribute file before the next promotional wave, so agents read the same facts twice
FieldWhat it holdsExample value
Arrival gatewayAssistant, search, feed, directAssistant
Mention typeNamed, shortlisted, linkedShortlisted
Basket valueOrder value in yuan428 yuan
Return rateReturns as a share of orders9.6 percent
Attribute completenessShare of required product fields filled86 percent

What a brand can actually measure

Three things are measurable now, and none of them is a ranking. Whether a session arrived from an assistant, which can be captured from referral parameters; whether the brand was named, shortlisted or linked, which has to be logged as three separate events; and how assistant-led baskets behave against browse-led ones on value and return rate. The rest, including which prompt produced the shortlist, is not visible to a seller this year.

The practical consequence is that product data becomes commercial infrastructure. An agent comparing supplements will read ingredient lists, serving sizes and certification fields, and a catalogue with empty attributes is invisible in a way that a poorly designed page never was. Tmall carried 1 million brands and 72 million products in this promotion; the number that will matter next year is how many of those products can be read correctly by a machine that has no patience for marketing copy.

FAQ

  • Is assistant shopping large enough to change a category plan?

    Not yet in aggregate, but already yes in specific categories. About 5 percent of Douyin purchases in October came through chatbot links, while supplements passed 20 percent. Where the buyer needs comparison help, the assistant sits between the brand and the sale.

  • What should a brand measure first?

    Whether the brand was named, shortlisted or linked, logged as three separate events, alongside the share of sessions arriving from an assistant. Those four numbers explain more than a ranking position, and they can be captured without platform cooperation beyond referral parameters.

  • Why did AI tokens become a promoted category?

    Because they are repeatable and measurable. Tmall reported token sales up more than 700 percent between October 15 and November 5, and a livestream sold 100,000 monthly QwenWork subscriptions in half an hour. The open question is renewal, not the launch price.

References