One in twenty purchases on Douyin now starts with a chatbot link. The figure, from WPIC Marketing + Technologies and reported as of November 11, 2026, looks small until you see the second half of it: for considered buys such as supplements, the same route accounts for more than 20 percent of orders. Set against a Singles Day that grew just 2.4 percent between October 7 and October 31 on Syntun's tracking, that split is the more interesting number for anyone who owns a product page.
A five percent channel that decides the shortlist
A chatbot does not scroll. It answers with three or four options and a reason attached to each. That changes what a product page has to win: not the click from a results page, but a place inside a sentence. Brands that live on price and stock availability can win that place; brands that live on a distinctive pack or a niche use case have to be described accurately enough in third-party sources for the model to repeat it.
The pattern shows up in the platform numbers. JD said its Dongdong assistant drew nearly 80 million users seeking advice this year, more than three times last year's figure. Alibaba reported that AI toys passed 100 million yuan in sales within a year of becoming a category, and that AI companion robots rose nearly fivefold year on year in the second half of October. These are not curiosity clicks; they are baskets assembled through a conversation, and the category growth they produce is concentrated where advice changes the decision.
Eighty million users are not browsing
Advice-seeking users behave differently from browsers. They arrive with a constraint, such as a narrow balcony, a quiet night mode or a specific budget, and they expect the assistant to filter. For a retailer that means the assistant becomes a merchandiser: it decides which SKUs are shown, in what order, and with which claim attached. For a brand, it means the page has to answer constrained questions explicitly, because a model can only repeat what it can find stated.
Considered purchases move first
Supplements crossed 20 percent because the category needs explanation: dosage, ingredient, suitability. Where explanation is valuable, an assistant earns its place, and where an assistant earns its place, the brand that supplies the cleanest facts gets quoted. Low-consideration categories will follow more slowly, mostly because shoppers do not need advice to buy a bottle of water.
Visibility becomes an auction
Ben Cavender of China Market Research Group put the risk plainly: brands can lose out if they stop appearing in product searches because shoppers let an assistant do the browsing. Platforms can tune what the assistant surfaces, which turns discovery into a shorter, more contested list where price, fees and paid placement carry more weight than brand memory. That is a structural shift in how a category leader is made.
Three discovery surfaces, three failure modes
Search results, assistant answers and livestream pitches are three different surfaces with three different failure modes. A team that measures only click-through rate will read all three as one funnel and misallocate the budget. The table below is the distinction worth writing into a weekly report.
| Surface | What decides inclusion | Typical failure |
|---|---|---|
| Search results | Keyword match and sales velocity | Ranking drops after a stockout |
| Assistant answer | Stated facts and citation quality | Wrong specification repeated |
| Livestream pitch | Host framing and price moment | Sales spike, returns follow |
The middle row is the one most catalogues are unprepared for. An assistant that repeats an outdated specification does damage that a ranking drop does not: the shopper is told the product is wrong for the job, in a sentence that sounds like advice. Keeping specification pages current is now a merchandising task, not a documentation chore.
What a brand should log this week
Start with a fixed set of questions, not a dashboard. Thirty prompts across brand, category and use-case phrasing, run at the same time each week, give a comparable series within a month. The metric that matters first is whether the brand is named at all, and second whether the sentence around it is accurate.
FAQ
Is a five percent share worth planning for?
Yes, because it is concentrated in considered categories where margins are higher. A category at 20 percent of orders through assistants is already a primary channel, not an experiment.
What makes a brand quotable in an assistant answer?
Stated, verifiable facts: specifications, compatibility, pricing, policy terms. A model repeats what it can find written plainly, and skips what is buried in images.
How is this different from search optimisation?
Search rewards keyword fit and sales velocity. Assistants reward clarity and source quality, because the answer cites something and the citation has to support the sentence.
Should assistant orders be reported separately?
Yes. Mixing them into search hides the fact that a different page element, the specification block, is doing the selling.
