Currently, most AI applications in the tourism industry are focused on pre-trip planning suggestions, which is considered the best application scenario for large-scale tourism models. As the leader in the OTA (Online Travel Agency) industry, Ctrip has launched its AI big model, Ctrip Asks, and Trip Genie, which were highly anticipated by many users. But how is the actual user experience?

A netizen from Heilongjiang: I asked Ctrip Asks to plan a trip from May 2nd to May 5th, and then asked it to recommend flights. It recommended a flight for May 3rd...

A netizen from Liaoning: I asked this AI about a three-day trip to Shenyang, and it said: "Day 1: Shenyang Palace Museum, Day 2: Zhang's Mansion, Day 3: Zhongjie Street." I immediately gave up! PS: These three attractions are all in the same area.

A netizen from Beijing: AI can't solve problems, but it can solve the people asking them. For example, after asking a series of questions about the best hotels in Shanghai, if you search for Ctrip hotels, they'll be more expensive! Because although it didn't solve your doubts, it grabbed your needs!

So, is the value that Ctrip Asks brings to users a "tool" or a "toy"? And in the context of various tourism AI big models emerging continuously, can Ctrip Asks maintain its claimed leading position?

Difficult to discern authenticity in planning

Amid the trend of major tech companies competing to develop and apply large models, smart cultural tourism vertical large models have also become popular. There are AI translation and AI tour guide capabilities provided by large models such as Wenxin One, Tongyi Qianwen, and iFLYTEK Spark from internet giants. There are also OTA-launched "Chengxin" big models, represented by Lvmama, which provide services such as intelligent itinerary planning, hotel reservations, transportation ticketing, attraction recommendations, and travel Q&A for tourists.

At the same time, many museums, attractions, and regions have launched AI companions, AI translation, and AI prediction products for the tourism direction, providing services such as exhibition consultation, exhibit navigation, recommended visit routes, and audio guides for tourists. Many tourists have already felt the ubiquitous presence of AI during their travels.

As early as when the AI big model "Ctrip Asks" was launched, James Liang, Chairman of the Board of Directors of Ctrip, publicly stated that solving accuracy issues would be key to AI service performance. "Tourism is a high-cost industry. Even if planning saves half an hour, there's a 5% chance that the recommended hotel or itinerary may be wrong, which isn't worth it."

In fact, users expect AI to replace or even surpass the functions of traditional travel agencies. They hope that when they don't know much about their destination city, AI can provide some usable and reasonable travel advice. Or in special situations like flight delays or traffic congestion, AI can play a role and help in complex itineraries.

Data and algorithms are crucial factors determining the AI user experience, but unfortunately, these two elements are not perceptible to users.

In many evaluations, AI big models are still dominated by human intervention, and users even need to spend extra time verifying the itineraries provided by AI. Taking Ctrip as an example, founded in 1999, Ctrip is obviously the company with the most accumulated user data in the online travel sector. Its data accuracy may be better than general data models that draw information from the entire internet, which can be heavily polluted. However, in terms of user semantic understanding, it is significantly inferior. This leads to users needing to spend more time for AI to understand their true intentions.

I did a trial run with Ctrip Asks: The first question was, "Please help me plan a trip from November 1st to 5th, with Yunnan as the destination." Ctrip Asks provided a 5-day itinerary with tourist attractions focused on Lijiang and Shangri-La.

Next, I asked another question, "Please recommend hotels in each place based on the above itinerary." Ctrip Asks recommended three hotels in Baoshan, Tengchong, and Lijiang. Since I knew nothing about Baoshan, I had to look it up online and found that it is a prefecture-level city in western Yunnan Province, bordering Dali Prefecture and Lincang City to the east, Nujiang Prefecture to the north, and Dehong Prefecture to the west. In other words, this hotel was neither in Lijiang nor Shangri-La.

I suspected there might be an error, so I asked the same question again, but the answer was exactly the same as before.

I asked Ctrip Asks to recommend flights for November 1st to 5th, and after asking three times, I got three different answers: The first time, it only recommended flights for the 4th; the second time, it recommended flights leaving on the 1st and returning on the 4th; and the third time, the same flights were recommended.

According to another evaluation by "Wenlv," during a conversation about a Shanghai itinerary, when the user asked about hotels near Zhongshan Park, Ctrip Asks provided accommodation information for all cities with Zhongshan Parks across the country, requiring the user to restate their destination.

Additionally, inaccurate restaurant operating conditions were a common issue reported by many users. Since users cannot ascertain the timeliness of the information cited by AI, if a restaurant they've chosen based on the AI's itinerary is closed, they have to spend extra time selecting another one. In terms of information timeliness, users cannot clearly distinguish which is more reliable: the AI's answer, travel videos, or guidebooks with photos and texts.

It's precisely this "inexplicability" from AI that leaves users unsure what data dimensions the AI's itinerary planning is based on. When AI repeatedly fails to clearly and accurately grasp the user's intentions, users will opt for more straightforward methods, such as "copying homework" – replicating other tourists' guidebooks with photos and texts.

Of course, the problems users encounter when using AI itinerary planning services are not unique to Ctrip. However, the disappointment lies in the fact that users expect more than just a search function with uncertain information sources from Ctrip's AI big model. As an industry veteran and leader, Ctrip should provide more precise and reliable functions and services. Unfortunately, Ctrip has failed to live up to this trust.

Users have different expectations for AI general and vertical big models. If the answers provided by vertical big models are even less accurate than those from general big models, what is the point of their existence?

Is the C-end tourism big model a paradox?

In its second-quarter report for 2024, Ctrip stated that in terms of itinerary planning, Ctrip Asks and Trip Genie can provide personalized, convenient, and efficient itinerary planning for global tourists. In terms of after-sales service, the application of AI and big models has significantly improved the self-resolution rate of customer complaints, saving nearly 10,000 hours of manual customer service work per day, equivalent to freeing up over 1,000 customer service personnel daily.

Freeing up 1,000 customer service representatives is undoubtedly a significant cost-saving measure and a notable highlight in the capital market. However, for users who have expectations for Ctrip's AI big model, they would like to see more specific data on how many users Ctrip Asks and Trip Genie have provided reliable services to in the area of itinerary planning and how much time and search costs they have saved.

At the same time, freeing up customer service workload can lead to another problem: when users need to resolve issues through human assistance, it becomes difficult to find genuine human customer service. A survey report released by Gartner shows that 64% of respondents do not want AI customer service. From the volume of complaints about customer service functions on various platforms, the cost for users to find human service has significantly increased.

The primary purpose of an excellent product is to address user pain points, especially in today's increasingly competitive and intense market.

Another example is that the original intention of AI itinerary planning is to address users' personalized needs, but does it really work? When users explicitly state in their questions that they want to avoid high-traffic attractions, the data used by Ctrip Asks precisely comes from "travel hotspots" and reputation rankings within the Ctrip ecosystem. The recommended content inevitably revolves around these highly popular attractions, failing to meet users' personalized needs. In such cases, AI big models need to differentiate between novice and experienced travelers and provide planning that is suitable and valuable for each group.

Extending from users' personalized needs, if they want to find less crowded attractions, they still need to self-identify through various social platforms. The information and data of AI big models are sourced from the internet, so is it a paradox to find less crowded attractions through an AI assistant?

Providing users with valuable personalized advice is one of the most challenging aspects of applying AI big models in the tourism industry. To overcome this challenge, tourism AI big models need continuous improvements, such as keeping abreast of the latest information, disclosing data sources when answering questions, and effectively screening data and miscellaneous information.

Vertical big models should inherently be more professional in their respective fields than general big models, providing richer and more valuable answers. However, if they merely offer search functionality, current tourism AI big models can hardly be considered products, as it's unclear what user problems they actually solve.

Users have grand visions for the future of tourism AI big models: issuing travel commands based on their needs and receiving replies that not only display essential information but also assist in completing recommendations and even bookings for flights, hotels, restaurants, and attractions, eliminating the need to switch between apps and mini-programs.

But judging from Ctrip Asks' current state, this goal is still far from being achieved!