One builds a closed ecosystem loop, while the other dominates search advertising. Doubao and ChatGPT ultimately find themselves on a converging path.

Written by | Landong Business Zhao Weiwei

Q: 'Find me the most cost-effective way to buy a 100-piece pack of genuine Doublemint gum, comparing prices on Tmall, Pinduoduo, and JD.com.' A: '...JD.com Arrow Brand Self-operated Flagship Store is our top recommendation; Pinduoduo offers the lowest prices with its 10 billion subsidies; Tmall Flagship Store/Tmall Supermarket provides stability and peace of mind... (Provides a purchase link for Douyin e-commerce, enabling one-click purchase without redirection).'

This was a straightforward Q&A example involving Doubao in late March. Activating product recommendation keywords prompts Doubao to display product links, allowing direct purchase from Douyin Mall without opening Douyin, even if the user's request is for other e-commerce platforms.

In AI application service-based transactions, Alibaba led the way; however, ByteDance has pioneered a true closed loop for in-app conversions in physical goods e-commerce.

During the same period, when the same purchase request was made to Alibaba's Qianwen, it couldn't directly provide Taobao purchase links. Qianwen explained that 'e-commerce platform commodity links are often time-sensitive, and directly providing external links may face expiration or security interception risks. To ensure accuracy, it is recommended to search and place orders on Tmall using keywords.'

Earlier, ChatGPT, which launched its advertising model in the United States, announced just six weeks after its debut that its annualized revenue, based on the current scale, had exceeded $100 million.

This is merely data from the U.S. market. With the self-service advertising system launched in April and the globally open advertising model, it is expected to generate $1 billion in revenue this year, with advertising revenue potentially nearing $25 billion by 2029.

AI commercialization has just begun. On the B-side, monetization occurs through MaaS (Model as a Service) and private deployment. On the C-side, profits are generated through subscription memberships, advertising, and e-commerce. As the AI-native products with the largest user bases in the Chinese and U.S. markets, Doubao and ChatGPT's C-side commercialization explorations represent the future direction of industry development.

ChatGPT is building its competitive edge through advertising, while Doubao is constructing a robust ecosystem within its own domain.

Whether Doubao directly sells goods to establish a closed e-commerce ecosystem loop or ChatGPT introduces an advertising model to capture search advertising markets, AI is transforming C-side e-commerce and advertising business models, ultimately converging on the same path. Advertising is ubiquitous, whether during retrieval or conversation. User intent and advertiser demands are precisely matched within a new commercial logic.

Doubao Achieves Closed Loop Before ChatGPT

During last year's Double 11, Doubao began integrating with Douyin Mall. When users inquired about product recommendations, Doubao would proactively suggest products, but users had to click a blue link to jump to Douyin for purchase. At that time, the functionality was relatively basic, and the user experience was not seamless.

By March this year, the Doubao + e-commerce feature had evolved. Users could complete the entire closed loop—from expressing needs, matching products, to payment and ordering—without leaving the Doubao App. Order information could be viewed within Doubao.

Technically, Doubao uses natural language to understand user consumption intent and then provides tiered recommendations based on Douyin Mall's product library, embedding product cards directly into the conversation flow. A single click from the user initiates the purchase process.

This means Doubao's commercialization closed loop for shopping guides may generate two profit sources: one is charging merchants promotion fees, similar to the ranking of previous search ads; the other is platform commissions on transaction volume, i.e., the CPS (Cost Per Sale) model.

Doubao's short-term GMV growth from traffic drainage and product recommendations may be limited, but most importantly, Doubao has achieved a complete closed loop for consumer decision-making within the App, realizing 'conversation is purchase' for product consumption. User data can flow seamlessly. If user trust is sufficiently high, it could structurally impact traditional e-commerce advertising, grassroots platforms, and price comparison websites.

In contrast, Douyin's local services have not streamlined the experience process as quickly as Douyin Mall.

When users ask Doubao for restaurant recommendations nearby, Doubao's responses remain unchanged. In these traditional recommendations, users need to click blue links and jump to Douyin to purchase group-buying discounts, resulting in less smooth local service consumption compared to product consumption within Douyin Mall.

Last year, ChatGPT started promoting products earlier than Doubao, driving user consumption decisions within the app. However, its prospects for completing the entire consumption closed loop are far less optimistic than Doubao's.

In late March this year, ChatGPT introduced a richer, more visually immersive shopping experience based on the Agentic Commerce Protocol (ACP). This evolution allows users to intuitively browse products, compare them side by side, and obtain real-time updated detailed information.

However, U.S. market users have not fully embraced this shopping guide feature. In October last year, a survey report from the University of Hamburg in Germany revealed that the conversion rate of traffic referred from ChatGPT was significantly lower than traditional marketing channels such as Google search, email, and affiliate links. ChatGPT brought more shoppers to e-commerce websites, but most did not make purchases.

Compared to the synergy between Doubao and Douyin Mall, OpenAI lacks its own e-commerce platform, payment infrastructure, and fulfillment system, relying solely on third-party connections. Therefore, ChatGPT's current product recommendation function serves more as a recommendation tool rather than a transaction closed loop.

ChatGPT's previously launched Instant Checkout feature has been suspended, with users now redirected to third-party retail websites for transactions. An OpenAI spokesperson stated that the company would deprioritize the independent development of Instant Checkout, instead prioritizing the development of consumer-facing product discovery features.

This means ChatGPT has retreated and contracted its e-commerce transactions, transforming into an intermediary for product consumption. It helps users decide what to buy through multidimensional data such as price, functionality, and reviews, serving as an entry point to the purchasing process rather than a platform for direct consumption on ChatGPT.

Advertising: OpenAI's Anxiety, Doubao's Gap

While ChatGPT has taken a step back in e-commerce closed loops, it has made significant strides in advertising commercialization.

Recently, OpenAI recruited Dave Dugan, Meta's senior advertising executive, as Vice President of Global Advertising Solutions. This move is interpreted by outsiders as strengthening OpenAI's connections with major advertisers. OpenAI urgently needs to open up new revenue sources to meet its vast artificial intelligence project and computing demands.

On the other hand, ChatGPT's advertising promotion business, which went live for trial operation for just two months this year and only appeared for free users in the United States, has already claimed an annualized advertising revenue exceeding $100 million, with over 600 cooperating advertisers. Internally, it is predicted that the business will generate approximately $1 billion in revenue this year, with advertising revenue potentially nearing $25 billion by 2029.

In tests conducted by Wired magazine, ChatGPT's advertisements appeared quite frequently, with roughly one ad every five conversational questions, accounting for about 20%.

ChatGPT's advertisements can now be customized individually. For example, when discussing flexible employment, an Uber ad appears, stating 'Your time, your income.' When asked about the worst TV shows in history, the AI provides 'The Jerry Springer Show' and 'Cop Rock' as two options, with an ad for Page Six Hollywood News Brief appearing below. When asked whether Harvard or Stanford is better, it triggers an ad for the University of Minnesota's part-time MBA program.

From the perspective of ad placement, ChatGPT's e-commerce ad recommendations, like Doubao's, are located at the bottom of the responses, not affecting the completeness of the AI's answers.

The trigger logic is also closely related to the topic. Each ad contains a website link button that redirects to the advertiser's external landing page when clicked. This follows the same logic as Google search ads, basically a hybrid model of CPC (Cost Per Click) + CPM (Cost Per Mille), resembling a parallel shift of Google search ads onto a conversational interface.

Currently, ChatGPT's advertising model is only promoted in the United States and only appears for users of the free version and Go subscription tier. According to OpenAI, only 85% of users actually qualify for ad display, and fewer than 20% of users see ads daily. OpenAI maintains caution and restraint in its advertising model, knowing that excessive ads could undermine users' trust in AI as a 'neutral advisor.'

With a minimal inventory of free users, OpenAI has achieved an annualized revenue of $100 million. The growth potential for OpenAI's advertising commercialization is immense.

In April, OpenAI will launch a self-service advertising tool and expand its market to Canada, Australia, and New Zealand, indicating an exponential surge in advertising revenue. However, controversy will also arise. Even though advertisers cannot influence the content of AI responses, such personalized ads may still make users feel monitored or that their privacy is being overly used.

'The worst thing a company can do is adopt an overly aggressive strategy that maximizes conversion rates and referrals but also damages people's confidence and trust in those recommendations. In that case, everything is over because there's no point in using a chatbot you don't trust,' Wired magazine quoted a professor from the Wharton School.

In late March, after securing the largest-ever funding round of $122 billion, OpenAI is currently valued at $852 billion, setting a new global record for AI enterprise valuations.

OpenAI currently generates approximately $2 billion in monthly revenue, with enterprise business accounting for 40%, expected to rise to 50% by the end of the year. Based on this estimation, user revenue, including advertising and membership subscriptions, is around $1.2 billion per month. Compared to the $100 million in annualized revenue from advertising, membership subscription revenue remains OpenAI's primary revenue source, accounting for approximately 99.3%.

Currently, the revenue growth rate from B-side enterprise business far exceeds that from C-side ad displays, which may be one reason why Doubao has not integrated third-party advertising promotions and instead reserves promotions solely for Douyin Mall.

The Hardest Part is Trust

Doubao and ChatGPT are on the same starting line, albeit with different approaches.

Whether Doubao directly promotes products to establish a closed e-commerce ecosystem loop or ChatGPT introduces an advertising model to capture search advertising markets, both point to a commercial endpoint: replacing search with conversation and transforming user intent into commercial value.

Their commonality lies in the fact that when users converse with AI, their needs are expressed more completely, authentically, and contextually than when entering keywords in a search box. When AI directly participates in transactions, shifting from traffic distributor to transaction facilitator, and potentially even becoming an AI commercial infrastructure with pricing power in the future, Doubao and ChatGPT, as leading AI products in the Chinese and U.S. markets, will gain the greatest first-mover advantage.

The difference is that Doubao relies on ByteDance's infrastructure in the e-commerce sector to take the lead in completing the closed loop for product transactions, while ChatGPT is still exploring the closed loop for e-commerce transactions and focuses more on the advertising intermediary model in the short term, aiming to become a high-value and highly concentrated advertising platform.

For ordinary users, the trust risks associated with AI are also intensifying. Whether users are willing to delegate purchasing decisions to AI and whether AI recommendations are perceived as soft advertising, leading to trust erosion, will affect user stickiness to AI products. User trust in AI products is the foundation for all commercialization efforts.

Currently, fierce commercialization competition among domestic large model vendors is more prevalent in the enterprise market. Whether through API calls or AI private deployment, cloud vendors are attracting enterprises to migrate or bind their entire IT architecture to their own public clouds by providing low-cost or even free high-quality foundational models. Large model vendors now compete not on who has the stronger model but on whose ecosystem and solutions can be implemented faster.

In contrast, the advertising model for AI applications is still in the experimental stage, with the AI advertising pricing system still in its formative phase and user trust still needing to be cultivated.

The only certainty is that when AI-native applications integrate advertising and e-commerce, traditional search advertising businesses will face the greatest impact. Users are accustomed to directly asking AI questions and receiving comprehensive answers, continuously weakening the role of search engines as information gateways. In this AI revolution, traditional search market share will be gradually eroded by AI platforms.