Author | Bishan

Source | Bowang Finance

Shortly after midnight on August 10, hotel merchants across the country were inundated with notifications about fee rates in their social media groups.

The notification originated from the Douyin Laike backend: Starting from midnight that day, hotel orders processed through the Doubao entrance and routed to Douyin Laike would be subject to a separate fee structure—11.4% for software service fees plus 0.6% for payment processing fees, amounting to a total comprehensive rate of 12%. Previously, orders generated by Doubao were lumped together with Douyin’s organic traffic and settled under a unified rate of approximately 8% for the accommodation category. Overnight, the channel cost for the same merchant and the same room surged by four percentage points.

That evening, Liu Xing, Doubao’s PR head, responded on Weibo, emphasizing two key points: First, Doubao’s lifestyle services currently do not involve paid promotions; hotels recommended by AI do not incur advertising fees, and merchants cannot influence recommendations or rankings through payments. Second, this fee is not a “recommendation fee” but a “channel service fee” charged only after a successful transaction—no transaction, no fee.

“It’s not an advertising fee; it’s a transaction service fee,” which is crucial to understanding the entire situation. This indicates that Doubao has not adopted Baidu’s old model of bid-based rankings—merchants do not need to bid for exposure, and user clicks are not charged. Instead, it aligns more closely with Ctrip and Meituan’s CPS (Cost Per Sale) commission model—the platform provides listing and transaction fulfillment, taking a percentage of the order amount. Mainstream OTAs typically have comprehensive commissions ranging from 10% to 15%, and when combined with bid-based advertising tools, merchants’ comprehensive channel costs can reach 15% or even exceed 20%. On its own, Doubao’s 12% rate falls in the middle of the industry range, with no upfront advertising costs.

However, a detail in the rules makes merchants more cautious. According to the platform’s announcement, the base for calculating Doubao’s software service fees is the amount actually paid by the user plus any official platform subsidies, while coupons issued by the merchants themselves are excluded. In other words, if the platform subsidizes 50 yuan to attract new users, this 50 yuan is also included in the commission base, with the merchant paying 12% of it—“platform subsidies, merchant-paid,” which differs from the long-standing industry practice of “charging based on actual revenue.” For small and medium-sized hotels with already thin profit margins, this calculation directly affects their willingness to invest in this new channel.

01 From “Natural Traffic” to “Transaction Channel”

What deserves more attention than the fee rate is the change in Doubao’s identity.

Previously, all orders directed to Douyin Laike by Doubao through AI Q&A were classified under Douyin’s natural traffic system—without independent identification or separate billing, essentially functioning as a free AI traffic tool within ByteDance’s ecosystem. On July 27, Douyin’s Lifestyle Services Learning Center released the “2026 Douyin Lifestyle Services Platform Software Service Fee Policy for Specific Channels,” listing Doubao as a “specific transaction channel.” Starting from August 10, Douyin, Doushengsheng, and Doubao will each have independent settlements, with group-buying service fees of 7.4%, 7.4%, and 11.4%, respectively. After adding a 0.6% payment processing fee, the final commission rates are 8%, 8%, and 12%.

This adjustment marks Doubao’s transition from a “free channel” to a “transaction channel”: independent settlement, exclusive pricing, and separate labeling of order sources. ByteDance has completed a full closed loop of “user AI consultation—personalized recommendations—redirected ordering—transaction-based billing.” According to Chen Zemin, an internet media analyst, Doubao’s commission on hotel merchants is more like a pilot test—using a high-precision AI traffic endpoint scenario for testing, with hotels being a niche industry that does not heavily overlap with Douyin’s main e-commerce site or core resources like Doushengsheng. The late-July policy also mentioned that “specific channels” could cover more lifestyle service categories. In other words, hotels are just the first test field, with dining, attraction tickets, local entertainment, and fitness likely to follow.

02 Doubao’s Accounts: Large Scale, Thin Revenue

To understand why Doubao is doing this, we must first look at its own financials.

The first page of the ledger is user scale. Data from third-party agency QuestMobile shows that in June this year, Doubao’s monthly active users reached 382 million, ranking first among domestic AI applications. This scale, in traditional internet terms, is almost like a money-printing machine—Taobao’s 900 million monthly actives correspond to annual revenue of about 500 billion yuan, Meituan’s 700 million monthly actives to about 300 billion yuan, and even Baidu, despite years of being written off, supports hundred-billion-level revenue with 600 million monthly actives.

The second page is thin revenue. According to LatePost, Doubao has over 200 million daily active users but generates less than 1 million yuan in daily revenue. As of June 2026, Doubao’s large model had a daily average of 180 trillion Token calls, growing over 1500 times since its launch—each dialogue and hotel recommendation involves daily inference costs in the tens of millions of yuan at the cloud level. With nearly 400 million monthly actives, traditional internet products have monetization efficiency tens or even hundreds of times higher than Doubao’s.

The third page is the collective pivot of the entire industry. On August 6, DeepSeek announced in its official API documentation a near-term overall price increase, “expected to be significant”; on August 10, Qianwen App officially launched a paid model; Zhipu had already raised API prices three times this year, with a cumulative increase of 83% in the first quarter; Yuezhi’s Dark Face released Kimi K3 with a new API pricing 3 to 4 times higher than the previous generation. The situation overseas is no less challenging: OpenAI reported a net loss of 38.5 billion USD in 2025, spending 2.6 USD for every 1 USD earned. ChatGPT has nearly 1 billion weekly actives, but its paid conversion rate is less than 10%.

Notably, there is price sensitivity among users. Third-party data shows that after Doubao announced its paid plans in May this year, its monthly actives decreased by about 6.07 million that month, a 1.81% month-over-month decline; in June and July, monthly actives fell by another 5.66 million and 6.92 million, respectively—a rare consecutive decline since its launch. Industry analyst Zhang Shule’s judgment is straightforward: “Relying solely on commissions to achieve an OTA model similar to Meituan and Ctrip is insufficient to cover Doubao’s R&D and operational costs.” In his view, this is more like a trial run for monetizing consumer scenarios, “or a risk exploration for future commercial use.” Another logic he provides is: “The habit-forming phase for large models among ordinary users is nearly over. Screening out truly high-engagement, high-demand users through charging is the incubation of a new round of consumer habits. The later the charging model is introduced, the more passive it becomes.”

03 Is 12% Expensive? Two Different Calculations

Returning to the most practical question: Is 12% expensive?

Within the industry’s framework, the answer is not simple. Based on 2024 GMV, the Ctrip ecosystem holds nearly 70% of the market share, with Ctrip’s main brand alone accounting for 56%; Tongcheng, Meituan, Fliggy, and Douyin have market shares of 15%, 13%, 8%, and 3%, respectively. The high-end hotel sector remains dominated by Ctrip, while Douyin’s strength lies in local lifestyle and homestay products. In this landscape, Doubao’s 12% is lower than the comprehensive costs of most leading OTAs after adding advertising, but higher than Douyin’s main site’s old rate of 8%—for merchants, whether it is expensive depends on the comparison.

Hotel merchants’ calculations are more nuanced. Doubao’s users are “people looking for goods”: they open a chatbox and ask for “cost-effective hotels near Beijing’s North Third Ring Road,” already carrying a clear booking intent, leading to fast decision-making and high conversion value. This is entirely different from the passive exposure of “goods finding people” while scrolling through short videos. If Doubao brings purely incremental customers, 12% is acceptable; but if it merely redirects natural traffic from Douyin’s main site to the Doubao channel for settlement at a higher rate, merchants are only buying existing traffic under a different name. Douyin’s lifestyle services hotel and travel GMV reached nearly 200 billion yuan in 2025, with a redemption rate of about 31%—the size of the existing traffic makes this question highly relevant.

04 The Account of Trust: Is AI the Answer or the Judge?

More difficult to calculate than fee rates is the account of trust.

The trust gap first arises from the changing role of AI. On Ctrip or Meituan’s interfaces, users see a scrollable, filterable hotel list with reviews after entering their criteria, with a relatively transparent display logic. In Doubao’s chatbox, however, AI recommendations are often simplified to three to five options or even a single “optimal solution.” When AI acts as a “decision filter,” its objectivity becomes the most sensitive issue—even if the official promises no bid-based rankings, once a commercial commission mechanism is established, doubts are hard to fully dispel.

Data confirms this caution. The “2026 China AI Travel Application Trend Insight Report” shows that over 60% of users still cross-verify prices, reviews, and services on traditional OTAs like Ctrip after receiving AI-recommended hotels, with only 15.2% placing orders directly. Media tests found that hotel booking prices via Doubao are indeed cheaper than on large booking platforms, but when a front desk employee at a Ji Hotel was asked about the price’s credibility, the advice was: “It’s best to book through platforms like Ctrip.”

The trust issue also hides a gray area. Zhao Jie, founder of Yanlin Cloud AI GEO, pointed out that even without explicitly priced advertising slots, merchants can still improve their chances of being mentioned and recommended by AI through generative engine optimization (GEO)—making brand content more structured and easier for models to understand. Moreover, models do not treat all sources equally: Volcano Engine’s June 2026 update to its “Site Authority Grading Guidelines” categorizes sources into four levels, with government websites and central media listed as the highest authority. Zhao also observed that all major models give some preference to their own content ecosystems—Doubao’s citations are shifting heavily toward Douyin video content, Yuanbao favors WeChat official accounts, Qianwen favors Taobao-related content, and Doubao favors Douyin. Whether this tilt represents ecological synergy or bias ultimately lies in the platform’s own interpretation.

05 Pricing Power at the Entrance

All of this ultimately points to the same question: Who gets to set the price at the intent entrance in the AI era?

In the traditional internet era, the answer was clear: Search boxes collected intent, and platforms auctioned that intent to suppliers. Ctrip is the intent entrance for travel, Meituan for food delivery, Taobao for shopping—the process of users repeatedly comparing prices gave platforms space to influence decisions, which was the source of advertising value. AI search, however, directly provides answers: intent remains strong, but user dwell time is only ten seconds, leaving minimal room for ads. The better AI performs and the more precise its recommendations, the harder it becomes to monetize its influence on decisions—this is the common commercialization paradox for all AI assistants.

ByteDance has provided its own answer: instead of selling ad slots, charge transaction fees. Industry data illustrates the potential of this approach—according to industry estimates, the global share of hotel bookings completed through AI agents rose from nearly zero two years ago to 3% to 5% in 2026, and is expected to exceed 8% by the end of 2027. When the transaction link is completed within a chatbox, the chatbot offers not just suggestions but also facilitates a purchase—this is essentially the same power that search engines and e-commerce platforms once held: determining what users see at the moment of decision-making.

Hotel merchants’ reactions will determine the direction of this trial. Platform officials admit that Doubao’s lifestyle services are still in an early exploratory and trial operation phase, with the core metric for evaluating channel value being the number of incremental orders and comprehensive customer acquisition cost, rather than the fee rate itself. For merchants, facing a new entrance with nearly 400 million monthly actives, the cost of abandoning it is equally high—especially in an era where overall hotel industry profits are being squeezed, and every new cost must be offset by incremental orders.

12% is a small number. The real big question is: When AI-provided answers also carry the platform’s financial interest, can users still trust that the three to five recommendations contain the “best” options, rather than the “most profitable” ones? The true price of this commission is not 12% but trust—it is the first thing Doubao must spend in every transaction going forward.