On August 10, Doubao began charging hotels a standalone channel fee of roughly 12% — 11.4% software service fee plus 0.6% payment processing — for bookings completed through its conversational entry point that redirects into Douyin's merchant system. Previously, those bookings were counted as Douyin natural traffic and settled at 8%. A footnote in a fee schedule, on its face. But it is the first time a conversational AI entry point has priced transaction traffic on its own terms — and that is the real story, not the percentage.
Doubao's shift from natural-traffic source to a dedicated settlement channel means a conversational AI entry has officially entered the online travel transaction chain. For the full analysis of what this signals for AI agents as emerging transaction hubs, see the companion piece Doubao's 12% commission: AI assistants are becoming new transaction hubs — same source, different angle: that piece is about how AI agents grow into transaction hubs, this one about how the OTA middleman model breaks.
The fee is the signal
The number matters less than what it proves: an AI chat surface can now set a toll for the traffic it generates, independently of the platform beneath it. ByteDance confirmed that judgment with a commercial act, not a press release. Doubao's ~382 million monthly active users make it the highest-frequency conversational surface in China, and it just converted that frequency into pricing power over hotel transactions.
The contrast with the incumbent is sharp. Ctrip, in its recently announced 19-item rectification plan, confirmed that its "AI business assistant" — launched in 2023 as a revenue optimizer that would help hotels price smartly — was fully taken offline in March. In practice it had degraded into a tool for Ctrip to control hotel pricing. So the old guard is dismantling its own AI instrument while the challenger openly charges for AI-mediated distribution.
Early data suggests this is a slow displacement, not a sudden one. The 2026 H1 China AI travel trends report found that 66.2% of users who receive an AI recommendation still return to an OTA platform to verify, while 15.2% trust the AI enough to buy directly. AI is not yet taking orders in bulk. What it is taking is the first decision position — the trust anchor. When a user verifies an AI's answer on an OTA, the OTA has already become the second stop.
What OTA actually sells is not service — it's information arbitrage
The profit base of traditional OTAs is not service; it is information arbitrage. Ctrip and Booking control inventory, prices, and review data that users cannot see, then sell access to that inventory through ranking and advertising — a toll road to the user. The elegance of the model is that users believe they are choosing freely while the platform's sorting algorithm has already decided what they can see.
AI conversation inverts this. On an OTA, a user says "find me a hotel" and receives a ranked list — the platform holds the decision. On Doubao, a user says "a place with a pool, close to the Forbidden City, quiet, under ¥800" and the AI understands the intent and returns an answer. A list can be ranked; an answer cannot. The search authority transfers from the platform to the user, and with it goes the ranking-based advertising that the OTA economy runs on.
The compounding effect is frequency asymmetry. OTA usage is planned and purpose-driven — the app opens only when a trip is certain. Doubao's usage is casual and ambient: users chat about where to go this weekend and the recommendation, and the booking, happen in the same flow. A 382M-MAU conversational surface attacks low-frequency transactional platforms with high-frequency everyday contact, and over time the review systems that gave OTAs their authority as decision references lose that authority.
Why this reads as a turning point
First, monetization is the proof of distribution. An AI dialogue entry has graduated from feature to channel the moment it can set its own toll. Once a channel prices traffic, it is a channel.
Second, the pattern generalizes. Any business whose moat is ranking plus information asymmetry — search, classifieds, marketplaces with opaque pricing — faces the same pressure: the answer replaces the list. The OTA is just the first industry where the shift is visible in a fee schedule.
Two caveats keep this honest. Trust migration is gradual: the 66.2% verification behavior shows the old infrastructure still holds the inventory and the receipts, and that is durable for now. And Doubao's move rides a closed loop that standalone assistants do not have — chat entry, Douyin merchant tools, and transaction settlement all under one roof. The flywheel is the moat, not the model alone.
What to do with this signal
For merchants and hotels: treat AI conversation channels as a distinct, paid distribution channel and audit your visibility inside AI answers, not just your ranking position. Channel costs will rise as AI surfaces monetize — plan for it.
For startups: the playbook is now on the table. Attach transaction capability to a high-frequency conversational surface, price distribution openly, and let intent-based answers do the marketing that rankings used to do.
For users: the rational pattern remains AI-first research plus OTA verification — the trust anchor is shifting, but the inventory still lives with the incumbents.
The 12% figure is small. The structural message is large: when an AI chat entry can set its own toll, the middleman model has found its successor. The open question is no longer whether AI will disintermediate the OTA — it is what an OTA becomes once it is no longer the gatekeeper.
AI agents are reshaping middlemen in more industries, a theme covered in the 2026 AI trends overview. On the Doubao fee itself, the authoritative report is Phoenix Guandian News: Doubao charges hotels a 12% commission - the first AI dialogue entry to do so.
