Doubao's e-commerce gamble is more complex than imagined.

Behind the daily average GMV of tens of millions, the AI shopping speed race has officially kicked off.

Yueji Kiln, Dehua, Fujian.

This is an ancient kiln with a 400-year history, and its signature offering is wood-fired white porcelain. In July this year, when the brand team checked the background data of their Douyin store, they noticed a new name in the order source column: Doubao.

No one paid much attention to it at first. But after pulling a full month of data, the team was quite surprised: from July to mid-August, the number of customer orders placed through Doubao increased by 10 times, with individual order values ranging from several hundred yuan to nearly 4,000 yuan, and the total transaction volume rose by about 677% year on year.

These orders came from conversations between users and Doubao, not from short video content marketing or live streaming rooms. Some users asked Doubao "what kind of tea set with high texture is suitable for gifting to elders", the AI recommended Yueji Kiln's white porcelain in its reply, and the user clicked the product card to complete the payment.

Yueji Kiln is not an isolated case. In the first half of this year, the daily average e-commerce GMV of the Doubao channel exceeded 10 million yuan. During the 618 shopping festival, the total GMV of the Doubao channel reached about 100 million yuan. Starting from July 15, Douyin E-commerce officially listed Doubao as an independent transaction channel, where merchants can separately view Doubao-generated orders, GMV and ROI data in their backend.

An experiment centered on "shopping entry" is quietly underway in the dialog box.

Three Tech Giants All Rush Into the Dialog Box Track

Looking back one year ago, "shopping in AI dialog boxes" was still a very niche concept. But in 2026, several major internet companies took action almost simultaneously.

ByteDance was the first to move. In October 2025, Doubao started preliminary integration with Douyin E-commerce, where users could get product recommendations and shopping links through conversations with Doubao. But at that time, users needed to jump to the Douyin APP to complete transactions, the experience was not smooth, and the feature was not rolled out on a large scale.

At the end of March 2026, Doubao officially launched the public beta of its "one-sentence shopping" feature. Users state their needs in the Doubao APP, for example, "I have a budget of 300 yuan for a pair of white running shoes suitable for commuting", the AI will push matched product cards, and users can place orders with one click without ever leaving Doubao. On April 24, the "Pick for You" function was officially launched on the navigation bar of the Doubao APP, realizing a fully closed loop of adding products to cart, payment and after-sales service all within the APP.

Alibaba followed closely. On May 11, Tongyi Qianwen announced full integration with Taobao, opening the entire product library of over 4 billion products on Taobao and Tmall to Qianwen. Users can complete product selection, comparison and order placement through natural language conversations in the Qianwen APP. The "Qianwen AI Shopping Assistant" was also launched synchronously in the Taobao APP, supporting functions such as AI virtual try-on, AI discount calculation, and AI low-price flash sale assistance.

Alibaba ran a large-scale test during the Spring Festival. The "Qianwen 3 Billion Yuan Free Order Campaign" was online for 6 days, with nearly 200 million "one-sentence order placement" operations, and 130 million people experienced AI shopping for the first time. This number made the entire industry realize that user acceptance of AI shopping may be higher than expected.

JD took the independent APP route. At the end of December 2024, JD launched the independent "JD AI Shopping" APP and opened its beta test, relying on its self-developed Yanxi large model to provide AI shopping guide services. Tencent connected its Yuanbao assistant with JD's AI Agent, allowing users to get JD product recommendations in Yuanbao, adopting the cooperation model of "AI assistant plus external e-commerce platform". Baidu's Wen Xiaoyan also started pushing JD product cards to users in May.

Xiaohongshu established Dots, a first-level AI department, at the end of April, and is also laying out AI shopping scenarios.

So many companies are doing the same thing at the same time, behind which there is a shared judgment: the traffic entry of e-commerce is about to usher in a major transformation.

In the past two decades, e-commerce entries have experienced several migrations. The earliest was the search box era, where users went to Taobao and JD with clear demands to search for products, which is called "people looking for goods". Then came the information flow era, where Pinduoduo and Douyin E-commerce used algorithm recommendation to push products to users, realizing "goods looking for people". Now large models have brought a third possibility: conversational shopping, where users describe their demands in natural language, and the AI understands, filters, compares prices and makes recommendations, realizing "AI helps people find goods".

The search box requires users to organize keywords by themselves, and the information flow requires users to make judgments in massive recommendations, while AI dialog can theoretically take over both of these two links. You don't need to know the specific model of the product, nor do you need to flip through reviews page by page. As long as you clearly state "who you are buying for, what scenario it is for, what your budget is, and what preferences you have", the AI can give a proper solution.

This is also why all three major platforms regard AI shopping as a strategic-level business. Whoever masters the shopping entry in the AI era will likely take the lead in the next round of e-commerce landscape.

Why Doubao's Closed-Loop Operation Runs Faster

In this AI shopping competition, Doubao's current commercialization data is relatively leading, which is related to ByteDance's unique resource endowments.

The first advantage is the user scale of the AI assistant. Doubao's monthly active users have reached 345 million, making it one of the largest AI assistants by user volume in China. A larger user base means more shopping demands can be tested and converted. On September 16 this year, the Doubao AI mobile phone was officially released, whose voice Agent can call all ByteDance's product lines. Saying "help me order a cup of coffee" will guide users to place orders through Douyin services, and the binding between the AI entry and transaction scenarios is further deepening.

The second and most critical advantage is the completeness of the closed loop.

Doubao and Douyin E-commerce both belong to the ByteDance system, with fully connected product libraries, payment systems and after-sales systems. Users do not need to jump out of the APP from conversation to payment. The technical link consists of intent recognition, demand parsing, product retrieval, sorting and recommendation, and in-APP payment. Some developers have calculated that the fastest time for users to complete the whole process from conversation to order placement through Doubao is only 15 seconds, while the traditional process of searching, comparing prices and jumping between platforms takes an average of 87 seconds.

Conversion rate data also proves this point. Data obtained by DoNews shows that the click conversion rate of Doubao AI-recommended products reaches 3%, which is higher than the average value of traditional information flow advertisements. In some local life scenarios, for example, when a tourist asks for "local dishes that are not spicy and suitable for non-local tastes", the AI recommends stores and guides users to visit them, and the conversion rate after click in some scenarios is even higher.

In contrast, competitors have encountered different degrees of problems in building a complete closed loop.

Although Alibaba's Qianwen has Taobao's 4 billion product library, it once voluntarily removed most product links in April. The Qianwen team explained that for categories with low standardization such as beauty products and apparel, links are prone to failure or point to unreliable stores, and "technical docking and resource integration have not been fully completed", so they did not want to hang links just for the sake of it. It was not until May 11 that they announced full integration.

This shows that even for e-commerce giants like Alibaba, it is technically difficult to deeply couple large models with massive product libraries.

Tencent's Yuanbao and Baidu's Wen Xiaoyan adopt the route of external connection with JD, which means the transaction link requires cross-APP jump, naturally adding extra friction to the user experience. JD's own independent AI Shopping APP faces the cold start problem, as users have not yet formed the habit of "opening the AI assistant first when shopping".

The third advantage of Doubao is the content data of Douyin E-commerce.

Traditional e-commerce platforms only know what users "bought", but Douyin also knows what users "watched, what they liked, and how long they stayed on certain content". The combination of these interest data and Doubao's conversation data can help the AI understand user preferences in a more three-dimensional way. For a user who often watches camping videos on Douyin and asks Doubao "what to bring for a weekend trip", the products recommended by the AI will be more in line with his actual interests.

Merchants are also starting to take this new channel seriously. After Doubao became an independent transaction channel on July 15, merchants can clearly see whether each order comes from Doubao in their backend. GEO (Generative Engine Optimization, which makes products easier to be recommended by AI) has become a new traffic operation topic following SEO and information flow advertising. Some operation service providers have already launched AI tools such as "Doubao Product Selection", teaching merchants to optimize product information to get AI recommendations.

In addition to physical product e-commerce, Doubao is also stepping into local life scenarios. Functions such as booking hotels and recommending restaurants through the Doubao entry have been launched one after another. However, this also caused a small controversy: in September, it was reported that Doubao charged a 11.4% software service fee plus a 0.6% payment processing fee for hotel orders, with a comprehensive rate of 12%, and the topic once topped the trending search list. The head of Doubao's public relations team responded the same day that "Doubao does not charge advertising fees for hotel recommendations", and the specific charging rules are subject to the platform's official announcement.

This small episode actually foreshadows the core issue of AI shopping commercialization: how much is the traffic brought by AI recommendations worth? When AI becomes the intermediate layer of consumer decision-making, how to distribute the benefits among platforms, merchants and users? All these rules are still in the exploratory stage.

Can You Really Buy Things in the Dialog Box?

Despite the rapid data growth, doubts about AI shopping have never stopped.

After reviewing various articles, these doubts point to a fundamental question: Is AI shopping a real demand, or a pseudo-demand created by large tech companies to tell AI stories?

From current practice, the answer may vary by category.

The first categories that AI shopping has successfully landed are those with low decision-making cost and high standardization, such as 3C digital products, daily groceries, and standardized food products. Users can clearly describe their demands ("phone case for iPhone 16", "diapers for two-year-old babies"), and the AI can quickly match suitable products, which is indeed more efficient than manual searching. The Qianwen team also mentioned that highly standardized categories such as 3C digital products are the first categories that AI shopping guides can stably bring links for.

Another scenario suitable for AI is product selection with "complex conditions but clear demands", for example, "sunscreen suitable for sensitive skin, within 200 yuan, for summer use", "shirt for 1.8m tall, 80kg weight, for business occasions". Such demands require repeated filtering and cross-comparison on traditional e-commerce platforms, while AI can complete multi-condition filtering in one conversation, saving a lot of time. A report from Xinhuanet mentioned that after the integration of Qianwen and Taobao, it mainly solves three types of pain points: cross-filtering with complex conditions, fuzzy intent reasoning, and scenario-based combined recommendation.

The real difficulty lies in non-standard products and experience-oriented consumption. The version, fabric and upper body effect of clothing, the size and style matching of home products, the skin feel of perfumes and skin care products, these highly subjective experience-dependent products are difficult for AI to recommend accurately only through text and images. The reason why Qianwen removed beauty and apparel links is exactly that the error rate of these categories remains high.

There is also a deeper problem: shopping is not all about efficiency, it also includes entertainment and social demands. Many people enjoy the process of "browsing" when they scroll Taobao and Douyin E-commerce, and the surprise of accidentally finding good products. When AI gives an optimal solution, it may deprive users of the fun of browsing. Just like navigation software can tell you the fastest route, but sometimes you just want to wander around casually.

Therefore, AI shopping is very unlikely to completely replace search and information flow. It will become the third shopping entry, coexisting with the other two. When users have clear goals, complex demands and want to save time, they will turn to AI for help; when they want to browse casually and discover new products, they will still scroll through information flow; when they already know exactly what they want to buy and only care about the price, they will use search. The three entries correspond to three different consumer mindsets, and share the same product library and transaction system behind them.

From this perspective, the combination of Doubao and Douyin E-commerce has a naturally complementary structure: the main Douyin APP satisfies the demand for "browsing" and "discovery", Doubao satisfies the demand for "consultation" and "decision-making", and Douyin Mall satisfies the demand for "search" and "repeat purchase".

At present, the proportion of GMV from the Doubao channel in Douyin E-commerce's total volume is still limited. The daily average GMV of 10 million yuan in the first half of 2026 translates to an annualized volume of about 3.6 billion yuan, accounting for less than 0.2% of Douyin E-commerce's half-year GMV of 2.4 trillion yuan. The GMV of about 100 million yuan during the 618 shopping festival is still a negligible fraction in the big promotion volume. It is more like a newly sprouted seedling, far from the harvest period.

But what the industry values is the trend and speed. From the preliminary integration in October last year to becoming an independent transaction channel in July this year, Doubao E-commerce only took 9 months. The 3% conversion rate of AI recommendation has exceeded the average value of information flow advertisements, which proves that "conversation as transaction" has been initially established in terms of technology and user experience. With the improvement of large models' product understanding capability and the maturity of multimodal interaction (identifying objects by taking photos, letting AI compare prices and place orders after watching marketing videos), the proportion of this channel may increase at a speed beyond expectations.

Several entry migrations in China's e-commerce industry over the past decade, namely search e-commerce, social e-commerce, and live streaming e-commerce, have each given birth to new giants and new brand opportunities. It is still too early to draw a conclusion whether conversational e-commerce will be the next one.

But it is certain that this time Chinese platforms and Silicon Valley giants are standing on the same starting line, and even run faster in landing speed. Amazon's Rufus and Shopify's AI assistant are still in small-scale tests, while Doubao and Qianwen are already running real transactions on a user base of hundreds of millions.

Whether the dialog box can become the next shopping homepage will be answered by the data in the next one or two years.

This article is from the WeChat official account "Xinmou" (ID: xinmouls), written by Tang Ning, and published with authorization from 36Kr.