“Fly from Guangzhou to Shanghai on April 2 to attend the TravelDaily Digital Intelligence Conference.”
A user typed this sentence into an AI travel assistant to see whether it could plan a business trip.
A few minutes later, the system proposed a plan: depart at 4 a.m. and arrive at the venue right at 9 a.m.
Thinking she hadn’t been clear enough, she added, “I can leave a day earlier.”
The AI quickly revised the plan—the departure time was still 4 a.m., only now the date had shifted to April 1.
By the time the full itinerary was generated, more than ten minutes had passed. In a real-world scenario, that would have been enough time for her to open an OTA app and book both flights and hotels.
Experiences like this are not uncommon. On social media, many users have shared similar frustrations when using AI to plan trips: illogical itineraries, inconsistent timing, or even routes that simply don’t make sense.
Despite the immature experience, user behavior is already beginning to shift. AI entering the travel industry is still widely seen as a major trend—it is changing how we travel and even reshaping the entire decision-making journey.
From “Search” to “Conversation” in Travel Decisions
For a long time, consumption on internet platforms has followed a “shelf-based” logic.
Whether on e-commerce platforms or OTAs, users typically enter keywords, browse through lists of products, compare options, and then make a purchase. The platform’s core capability lies in standardizing product features and presenting them to users, so they can make their own selections.
But in the age of AI, this path is being rewritten.
According to Ye Chen, CTO of Fliggy, future travel decisions may no longer start with search, but with exploration. Users gain inspiration from content, form ideas, and then move into planning, booking, and fulfillment.
They begin with inspiration, develop intent, move into itinerary planning and research, complete transactions, and ultimately fulfill the trip.
At the same time, the way platforms understand users is also evolving.
“In the past, users could only express their needs with two or three keywords—that was a limitation of product design,” Chen said. “Now, users can express much richer intent through conversation.”
“Instead of searching and filtering, users start by chatting,” he added. “After a short conversation, the system can generate a highly personalized plan.”
However, there is still a significant gap between “a smooth conversation” and “a smooth journey.”
Why AI Hasn’t Taken Over Travel Yet
Based on current experience, AI is still far from taking over the full travel chain.
The core issue is that general-purpose models lack structural capabilities when dealing with the complexity of the physical world.
First, there is a lack of temporal and spatial common sense. “For humans, it’s easy to recognize that a midnight departure is unreasonable,” Chen said. “But for a general model, that judgment isn’t innate.”
Second, there is insufficient ability to handle real-world constraints. Flights, transportation, and attraction schedules are tightly interdependent—any deviation can break the entire plan.
Chen gave an example: a user wants to leave Hangzhou at 7 a.m. and arrive at Shanghai Disneyland by 9:30 a.m. The model suggests a seemingly perfect plan—7 a.m. high-speed rail, arriving at Hongqiao at 8 a.m., followed by a transfer. But in reality, rush-hour congestion, transfer time, and park opening uncertainties make the plan unworkable from the start.
More critically, there is the ability to verify against reality. “The model must not only plan, but also validate,” he emphasized. “It needs to call external data—maps, timetables—to check feasibility.”
Even if these issues are gradually solved, a deeper constraint remains:
Travel is not a scenario where trial and error is acceptable. It is a fulfillment process that must succeed the first time. When users need not just a plausible plan, but a reliable execution outcome, the challenge is no longer about generation—it becomes a systems problem.
Still, Chen believes this is only a matter of time. “Once this threshold is crossed, AI will evolve from a supporting tool into a new service entry point.”
AI Could Become the New Travel Entry Point
If AI becomes a super entry point, will users still need OTAs?
Chen believes interaction models could change completely. “In current stage, we believe in the concept of a super entry point,” he said. “In the AI era, there will be an AI-native super interface—perhaps chat-based, or something more advanced.”
This entry point will no longer be limited to a single app. It will exist across devices—phones, wearables, cars—sharing a unified memory of the user. When a user gives a vague instruction like, “After my afternoon meeting, go to Shanghai and be back home in time for dinner,” the assistant can orchestrate resources across the backend, completing everything from planning to execution.
“Users won’t need to describe every detail, because this entry point understands them better than an assistant,” Chen said.
As such an AI interface takes shape, the underlying supply system must also reorganize around it. Flights, hotels, and transportation services must become directly callable by AI. This implies that the role of traditional OTAs is evolving.
In the internet era, OTAs created value by connecting supply and demand through its search and price comparison system. In the AI era, access to services may shift from “searching for products” to “talking to AI.” “A model that understands users can arrive at the optimal solution through just a few exchanges, replacing the need for manual filtering,” Chen said.
At the same time, fulfillment assurance may become even more important. Travel is full of uncertainty—delays, cancellations, missed connections. If AI can coordinate backend systems in real time and proactively resolve disruptions, platforms could evolve into true “travel concierges.”
The supply chain itself also needs to be upgraded for AI. Current product descriptions, based on a few dozen structured fields, cannot satisfy fine-grained requests like “a hot spring hotel with a view of Mount Fuji.” The boundaries and responsibilities of OTAs will need to be redefined.
AI Changes the Interface, Not the Platform
In recent years, there has been a persistent concern: if AI assistants become the main entry point, will users bypass OTAs and book directly with airlines or hotels?
Recently, however, sentiment has begun to shift.
Reports suggest that OpenAI has slowed down efforts to embed direct booking into ChatGPT, instead emphasizing transactions through third-party applications. This indicates that platforms like Booking and Expedia are not facing immediate existential threats.
From an industry perspective, the concern itself may be misplaced.
The idea of “bypassing platforms” is not new. Airlines and hotel groups have long used search advertising to drive direct bookings. Yet this model has never become dominant. The reason is simple: when users must switch between systems, re-enter information, and bear transaction risks themselves, the experience becomes worse.
In Chen’s view, some fundamentals of the OTA business model will remain unchanged.
“Today, OTAs’ revenue mainly comes from commissions and advertising,” he said. “In the AI era, this logic will still hold—only the attribution and the format may change, for example, shifting from search ranking fees to API call fees.”
Another enduring element is membership systems. Building long-term relationships and addressing retention through membership predates the internet.
But this does not mean stagnation. In Chen Ye’s view, if OTAs are to move into the future, their role should grow. They must not only connect supply and demand, but also empower the entire industry adapts to the AI era—so that products can be more naturally discovered and consumed.
This shift will ultimately redefine roles.
When you say to a super interface, “I need to go to Shanghai this afternoon and be back before dinner,” AI is responsible for understanding the request. The platform is responsible for ensure the process of turning it into a real ticket, an on-time journey, and a dinner you don’t miss.
Above the interface is AI. Beneath it is the platform. AI takes the entry point, but the platform still controls the outcome.
The supply chain and fulfillment capabilities built over the past two decades remain the hardest part of the travel industry to replace. AI changes the interface—but not the underlying structure.
In this new cycle, those who build deeper beneath the interface will stand stronger.
