There is a class of China AI visibility signal that no amount of content work will move, and in 2026 it quietly became the most important one for consumer brands.
When a shopper asks Qwen for a humidifier under 300 yuan, Qwen does not compose a prose recommendation from crawled articles. It returns order cards drawn from Taobao's roughly 4-billion-SKU catalogue, and those cards compare price, sales volume, rating, and store tier. When a shopper asks Doubao's "帮你选" ("Pick For You") feature the same question, Doubao resolves it against Douyin Mall, where Chinese trade press now describes the merchant Experience Score (体验分) as the effective gate on AI traffic allocation, product surfacing, and store weighting.
Neither path reads your brand story. Both read your operations ledger.
This is a structural break from the GEO playbook most international brands imported from the West. In a citation-based answer, a well-sourced Zhihu post or a Baidu Baike entry can lift you. In an agentic commerce answer, the model is not synthesizing opinion — it is filtering a structured product database by fields you populate through fulfillment performance, dispute rates, and review velocity. A brand can hold a strong GEO score on informational queries and still be invisible at the moment of purchase.
Two commerce loops, two signal stacks
May 2026 closed both loops within weeks of each other. Qwen went live with full Taobao integration — described by Chinese outlets as the first end-to-end AI shopping loop, spanning intent understanding, recommendation, ordering, fulfillment, and after-sales. Doubao, which had already connected to Douyin Mall in October 2025, launched 帮你选 and completed its own full-chain integration with Douyin e-commerce.
The two loops inherit their parents' ranking logic, and the differences matter for where you spend.
| Qwen × Taobao | Doubao × Douyin Mall | |
|---|---|---|
| Catalogue basis | ~4B SKUs, 20+ years of transaction data | Douyin Mall inventory, content-native |
| Surfaced fields | Price, sales volume, rating, store tier | Experience Score-weighted surfacing |
| Query type it wins | Shelf-style, specification-led, comparison | Discovery-led, lifestyle and impulse |
| Brand lever | Store tier, review depth, price competitiveness | Experience Score components, content-commerce linkage |
| What content buys you | Entity clarity upstream of the card | Discovery reach into the mall |
Read the table as a budget allocation instrument. The final column is the one most brands over-fund. Content buys you consideration — it does not buy you a position on the card.
Why the Experience Score deserves a line in your GEO budget
Douyin's Experience Score is not new; it has governed merchant traffic for years, composed of product experience, logistics experience, and service experience sub-scores. What changed in 2026 is its scope. Chinese industry coverage frames it as moving from a traffic-allocation input to the assessment standard behind AI-side recommendation and store weighting — which is to say, the score now propagates into Doubao's answers.
For an international brand, that has three uncomfortable implications.
First, the signal is owned by whoever operates your Douyin store. If that is a TP (Tmall/Douyin Partner) agency or a distributor, your AI visibility in China's largest consumer assistant is being determined by a vendor whose contract almost certainly does not mention AI visibility. Doubao's reach makes this consequential: reported user figures range from 330 million by May 2026 to claims of 600 million MAU in domestic marketing coverage, a spread wide enough to warrant caution about any single number, but not wide enough to make the platform optional.
Second, it is slow to repair. A brand can publish a Baike revision and see citation effects within weeks. Experience Score components are trailing averages over transaction windows — a bad quarter of logistics performance depresses your AI surfacing for months after the operational problem is fixed. This is the longest-latency GEO signal in the Chinese stack.
Third, it does not appear in any GEO tool's output, including score-based trackers like ours. Brand score measurement samples model answers to informational and recommendation prompts across Doubao, Kimi, DeepSeek, Qwen, Wenxin, and Hunyuan. It captures whether a model names you. It does not capture whether the commerce layer beneath the model will put you on a card. Two brands with identical scores on hubGEO's brands index can have completely divergent purchase-moment visibility.
The paid-ranking question nobody has answered
Chinese tech commentary on the Qwen × Taobao launch raised the obvious suspicion immediately: is this genuinely quality-ranked, or is it bid ranking wearing a conversational interface? The honest answer is that it is not externally verifiable. The order cards expose price, sales, rating, and store tier — all plausibly organic — but the weighting function is undisclosed, and Alibaba's incentive to monetize the highest-intent surface it has ever built is not in question.
We flagged the beginning of this shift when Doubao started rolling out paid placement. The commerce loops raise the stakes, because a paid slot inside an AI shopping card is not an ad adjacent to a result — it is the result. Brands should plan on the assumption that both loops introduce explicit paid ranking within the next four quarters, and that the current window of operationally-earned visibility is temporary.
The models with no commerce loop are not off the hook
Kimi, DeepSeek, Wenxin, and Hunyuan have no equivalent closed purchase loop today. It is tempting to conclude that the operational-signal argument does not apply to them — and for card placement, it does not. But the second-order effect is worth naming.
Commerce loops generate a new class of training and retrieval signal: structured records of what shoppers actually selected after seeing a comparison. That data stays inside ByteDance and Alibaba. Over time it widens the quality gap between Doubao's and Qwen's product recommendations and everyone else's, which pushes purchase-intent query share toward the two loop owners regardless of headline MAU.
For brands, the planning implication is that platform prioritization should be split by intent rather than by audience size. Doubao and Qwen deserve the transactional budget because they own the purchase moment. DeepSeek and Kimi still deserve informational investment, because their users skew toward research-heavy and professional audiences whose consideration happens before any card is rendered — which is precisely where entity clarity and citation quality do their work. Hunyuan's position inside WeChat makes it the wildcard: a commerce loop there would be the largest of the three, and Tencent has the pieces.
What this changes about measurement
The practical consequence is that "China AI visibility" needs to be split into two tracked surfaces rather than one.
The informational surface is what conventional GEO measurement covers: does the model know you, describe you accurately, and recommend you when asked an open question. Levers are entity consensus, authoritative citations, and content freshness. Latency from action to effect: weeks.
The transactional surface is the commerce loop: will the agent put your SKU on the card. Levers are store tier, Experience Score components, review depth, price position, and fulfillment reliability. Latency: months, and the lever often sits with a third party.
Most brand teams have an owner for the first and nobody for the second, because the second looks like an e-commerce operations problem rather than a marketing one. Organizationally, that is exactly the gap that will get exploited when paid ranking arrives — the team that would have negotiated placement is not the team that watches the surface.
Takeaway for Brand Marketers
Do four things this quarter.
Audit who controls your Experience Score. Identify the entity operating your Douyin storefront, pull the current product, logistics, and service sub-scores, and get them into your marketing dashboard. If a TP agency holds them, add AI-surfacing performance to the next contract review. This is the single highest-leverage move available, and it costs nothing but internal coordination.
Instrument the transactional surface separately. Run recommendation prompts through Doubao's 帮你选 and Qwen's shopping flow monthly and record whether your SKUs appear on cards — not just whether the model mentions your brand in prose. Treat divergence between the two as a diagnostic, not noise.
Rebalance away from pure content spend. If your China GEO budget is entirely content and citation work, you are optimizing one of two surfaces. A reasonable starting split for consumer brands with a transactional presence is to move a meaningful share toward review depth, fulfillment performance, and store-tier improvement.
Price in the paid-ranking transition. Build your 2027 China plan on the assumption that AI shopping card placement becomes purchasable. Brands that enter that auction with strong organic operational signals will pay less for the same position — which makes Experience Score work an investment in future media efficiency, not just current visibility.
The brands that lose the next two years in China will not be the ones that failed to write enough content. They will be the ones whose AI visibility was quietly delegated to a logistics vendor.
Related: see current model-by-model brand scores on the brands index, and our earlier analysis of Doubao's paid ad rollout and agentic checkout as a silent brand selection layer.
Sources: TMTPost and Infoobs on the two AI commerce routes (May 2026); Yangtze Evening Post on the Qwen × Taobao launch; China.com / Sina Tech on Doubao, Douyin stores, and the Experience Score; Yesky commentary on paid ranking in Qwen shopping results; CNBC and Global Times on Chinese agentic commerce.