Alibaba and Amazon: Two Differentiated Samples of AI E-commerce

In recent years, AI companies have been trying to transform the e-commerce industry but have remained at the stage of "recommending products to users." It wasn't until recently that two global e-commerce giants took a crucial step almost simultaneously.

On May 11, just before the 618 shopping festival, Alibaba announced the full integration of Qianwen with Taobao. Users can now directly complete product selection, price comparison, order placement, and after-sales service through AI. Two days later, Amazon also announced the upgrade of its AI shopping system and launched a new shopping assistant, Alexa for Shopping, to strengthen the closed-loop ability from search to transaction.

This is not simply an "upgrade of AI shopping guides." More and more technology companies are beginning to realize that whoever controls the AI entry point may redefine the traffic distribution system of the next generation of e-commerce.

In the past two years, global technology companies have engaged in a fierce competition around the question of "whether AI can take over the shopping entry point." AI-native companies such as OpenAI and Google have tried to reconstruct consumer decision-making with chatbots. Traditional e-commerce platforms led by Amazon hope to embed AI capabilities into their own ecosystems to prevent users' entry points from being intercepted by AI.

E-commerce is becoming the next core battlefield for AI Agents.

AI Shopping: Moving Towards Three Camps

After nearly three decades of development in e-commerce, although users' shopping paths have evolved from single search to search + recommendation, the essence of manual operations, including browsing products, comparing prices repeatedly, and then placing orders, remains unchanged. Facing a vast supply of products, users often need to spend a lot of time screening information, confirming needs, and may even need to return or exchange products multiple times to find the truly suitable ones.

What AI is trying to change is precisely this process.

Compared with traditional search, AI can understand vague needs, integrate product information, and automatically complete screening, recommendation, and price comparison based on user preferences. From "helping users find products" to further taking over the transaction process, AI is redefining the entry form of e-commerce.

Due to differences in business models and technological capabilities, global technology companies have gradually formed three different camps.

The first camp consists of traditional e-commerce companies led by Amazon. They usually embed self-developed AI assistants into their platforms and provide all necessary functions from search to order placement based on their own large product databases. However, their large model reasoning abilities are generally considered to be inferior to those of AI-native companies.

Take Amazon's first AI assistant, Rufus, as an example. Rufus can only be used on Amazon's website and app. It can conduct conversations, compare prices, and check reviews. By 2025, it will have served over 300 million users. However, media evaluations suggest that Rufus often "answers off-topic" and "misses the point" in complex product comparison scenarios.

On May 13, Amazon announced the integration of Rufus into the Alexa+ system and launched Alexa for Shopping. This new AI shopping assistant expands the shopping scenario to smart devices such as Amazon's Echo Show. It not only supports intelligent search, price drop alerts, product comparison, and automatic reordering but also helps users select products on websites outside Amazon and track logistics.

In this way, the originally scattered actions such as search, screening, price comparison, and order placement are centralized within the Amazon ecosystem, further strengthening the retail closed-loop of this e-commerce giant.

The second camp consists of AI-native companies such as OpenAI and Google. They have top-notch large model reasoning abilities and can provide the most accurate product recommendations. However, they lack infrastructure such as logistics, payment, and after-sales service, so most of them stop at "recommendation - redirection."

In April last year, OpenAI built a shopping function into ChatGPT. After users enter instructions in the chat box, ChatGPT will display product reviews and purchase links from e-commerce platforms such as Walmart and Best Buy based on historical conversations and browsing preferences. Users can click the purchase button to jump to the corresponding platform for further operations.

Google's Gemini also provides a similar AI shopping service, allowing users to "complete transactions within the dialog box." However, Gemini has a more diverse range of linked products - in addition to "external resource libraries" such as Walmart and Shopify, it also has 45 billion structured product data accumulated by Google Shopping over the years.

That is to say, while these companies are seizing the traffic entry point that traditional e-commerce companies value most, they must rely on traditional e-commerce or payment platforms to complete the shopping closed-loop, and these two are naturally contradictory.

The companies in the third camp choose to "walk on two legs" - deeply integrate ultra-large-scale e-commerce platforms with top-tier large model applications. Alibaba's full integration of Qianwen with Taobao is a representative case of such companies.

Specifically, after integrating with Taobao, Qianwen can accurately understand users' consumption intentions in conversations based on Taobao's 4 billion product library and real shopping scenario data accumulated over more than 20 years, and provide precise product recommendations to help users make shopping decisions in complex environments. After Taobao embeds Qianwen, it will launch an "AI shopping assistant" to provide users with various AI shopping functions based on real pain points in different shopping scenarios.

Alibaba claims that Qianwen has achieved the first full-process closed-loop of AI shopping in the industry, from demand understanding, product recommendation, to order placement, fulfillment, and after-sales service. Taobao has also taken a crucial step towards the AI transformation of a super app.

Why Can't AI Companies Complete the Shopping Closed-Loop?

The three camps may seem clearly defined, but in fact, they are intertwined and full of undercurrents.

To make up for the shortcomings in transaction scenarios and fulfillment capabilities, AI-native companies are trying every means to obtain product data resources from external e-commerce platforms, attempting to retain users with the "one-sentence shopping" experience. This move undoubtedly invades the territory of traditional e-commerce companies.

Once users get used to the AI chat entry point, the status of traditional e-commerce platforms as information and shopping entry points will be severely impacted. Amazon realized this early on and fired the first shot in the "Agent counterattack."

The cause was the newly launched browser Comet by the US AI startup Perplexity. Different from ordinary browsers, Comet itself is an AI Agent. After obtaining user authorization, it can disguise as a Chrome user, log in to accounts, and complete shopping on behalf of real people.

In November last year, Amazon sued Perplexity, accusing Comet of computer fraud for its behavior of proxying consumers to shop on Amazon's website. The reason was that it did not clearly disclose its actions when proxying users to shop, violating Amazon's service terms, and refused to stop the behavior when Amazon requested it.

In March this year, the San Francisco Federal Court issued a temporary injunction, stating that "although it has obtained user permission, it has not obtained Amazon's authorization," requiring Perplexity to stop accessing Amazon users' accounts and destroy relevant data copies. Previously, Perplexity publicly stated that Amazon was using its own competing products to suppress smaller competitors, and users should have the right to choose their preferred AI shopping agents.

While keeping traditional e-commerce companies on the defensive, AI-native companies also have their own troubles.

In September last year, OpenAI launched the "Instant Checkout" function in cooperation with payment platforms such as PayPal and Shopify, allowing users to place orders and pay without jumping to third-party websites. However, this function was stopped after only six months due to low conversion rates.

OpenAI's internal data shows that although a large number of users browse and compare prices in ChatGPT, almost no one actually places an order in the chat interface. According to data from the partner Walmart, the conversion rate of returning to the retailer's official website to check out is three times that of staying in ChatGPT.

The cold chat box cannot give users the sense of security they want.

A report from Morgan Stanley pointed out that among the respondents who refused to use AI shopping, 36% did not trust AI recommendations, 34% preferred to search for and purchase products on retailers' websites, and 31% were worried about privacy and personal data security.

An e-commerce practitioner told the media that model companies cannot obtain the most core data of e-commerce platforms, and e-commerce infrastructure cannot be fully opened to external companies. There is always a wall between the two companies, and all AI can do is "recommend + redirect," making it difficult to go deeper.

However, no company is willing to "divert traffic" for other platforms forever. With limited closed-loop capabilities, AI-native companies are starting to explore new transaction paths.

Different from consumer-oriented AI shopping, some companies are starting to try a more radical "agent economy" direction. A recent internal experiment by Anthropic shows that Claude can independently complete negotiations and transactions in a real market environment. Anthropic believes that this means that a new business form of direct negotiation, price inquiry, and transaction between AIs is emerging.

What Is Alibaba Betting On?

Compared with overseas technology companies, Chinese companies are advancing faster in commercialization and scenario application. Whether it is the population size, industrial structure, or users' acceptance of new technologies, they provide a natural test field for AI products to be quickly verified and iterated in more complex and diverse real environments.

Alibaba is one of the most comprehensively deployed Internet companies in the field of AI shopping - it has an e-commerce "gene," a first-class large model, and one of the largest third-party payment platforms in China.

During the Spring Festival this year, Qianwen invested 3 billion yuan in the "Spring Festival Treat Plan," integrating with Alibaba's ecological businesses such as Taobao Flash Sale, Fliggy, Damai, Hema, Tmall Supermarket, and Alipay to encourage users to place orders through Qianwen. Now, the integration of Taobao fills an important gap in AI e-commerce and helps Alibaba build an irreplicable business barrier.

As Qianwen is deeply integrated with Taobao, a new question is emerging: Why does Alibaba promote both "native AI + e-commerce" and "existing e-commerce + AI"? What is the essential difference between the two?

In the "native AI + e-commerce" shopping system centered around the Qianwen App, users can directly complete product search, comparison, order placement, and even after-sales service through natural language, without relying on traditional keyword searches. Qianwen can also call on Taobao's logistics, fulfillment, and after-sales capabilities to truly achieve "agent shopping."

This means that Qianwen's goal is not only to improve the shopping experience but also to try to take over the consumption decision-making entry point. In the future, users may only need to describe their needs to Qianwen, and AI will automatically complete screening, comparison, and transactions. From a business structure perspective, AI Agents will weaken the most core asset of traditional e-commerce platforms - traffic distribution rights.

The integration of Qianwen into the Taobao App can be regarded as an AI upgrade for Taobao itself. The currently announced capabilities include AI virtual try-on, AI discount calculation, and AI low-price assistance, which essentially still serve Taobao's original platform logic. In other words, the integration of Qianwen into Taobao further strengthens the existing shelf e-commerce system.

The biggest difference between these two paths lies in their different attitudes towards the "consumption process."

Traditional e-commerce relies on users' stay time, content consumption, and advertising exposure, so it naturally encourages "browsing." The longer users browse, the easier it is for the platform to complete recommendations, "planting grass," and advertising monetization. Therefore, even when introducing AI, Taobao is more inclined to use AI to improve search efficiency and conversion rates rather than completely compress the consumption process.

However, Qianwen's logic is exactly the opposite. The core goal of AI Agents is to complete tasks in the shortest path. In the future, users do not need to browse a large number of products, compare parameters repeatedly, or even actively open e-commerce platforms. Instead, they can directly submit their needs to AI, and AI will complete screening, decision-making, and execution.

This is why Alibaba must promote both the "native AI + e-commerce" and "existing e-commerce + AI" routes.

On the one hand, it needs to safeguard Taobao's existing advertising, merchant, and transaction systems. On the other hand, it must compete for the next-generation AI entry point in advance to prevent users from directly making consumption decisions through AI in the future, thereby bypassing traditional platforms.

In the past two decades, e-commerce platforms have controlled the "product entry point": users search, browse, compare prices, and then complete transactions. The core competitiveness of the platform is traffic distribution ability. However, after the emergence of AI Agents, the entry point has started to shift from "platform pages" to "dialog boxes" for the first time.

This means that in the future, the most important asset of e-commerce platforms may no longer be homepage traffic but who can influence AI recommendation results. The competition among merchants will also shift from "search rankings" to "AI recommendation rights."

The most fundamental distribution logic of e-commerce is being reconstructed by AI. And the answer to who will become the next-generation consumption entry point remains to be seen.

This article is from the WeChat official account "New Berry Daybreak" (ID: new-daybreak), written by Wang Mumu and published by 36Kr with authorization.