Older travellers were more likely to adopt an AI shopping assistant on Ctrip, according to research using records from more than 31 million users. The March 2026 preprint also found higher adoption among women and established, highly engaged customers. The result challenges a familiar technology stereotype, but its scope is specific: one assistant embedded in a Chinese travel platform, rather than AI use across society.
The useful question is why a shopping feature might attract a different audience from a standalone chatbot. Somebody need not be interested in artificial intelligence as a subject to find help planning a trip worthwhile. Adoption can be understood through the task a person already wants to complete, rather than their enthusiasm for a new technology category.
An AI shopping assistant inside an existing habit
The researchers studied Wendao, Ctrip’s conversational assistant. Their base sample comprised 31,142,353 users who logged into the app between 10 and 24 July 2025. Adoption meant having initiated at least one assistant chat by 10 July, which applied to 1,904,368 users, or 6.1%. The paper therefore describes behaviour, not a survey of intentions.
That definition matters. Trying something once, returning regularly and trusting it with a purchase are different stages. A platform can record the first without establishing the other two. Readers should not translate an adoption percentage into a claim about how many people delegated their holiday plans or became confident AI users.
Imagine a customer already using an app to compare rail tickets and hotels. Asking an additional question inside that familiar setting is a smaller change than choosing a separate AI service, opening an account and working out how to use it. This is a plausible explanation for why context matters, not a causal mechanism established by the Ctrip data.
The assistant did not simply replace search
The study found users moving between chat and conventional search, rather than following a universal chat-to-checkout path. Attraction questions represented 42% of observed chat requests. The researchers interpret the pattern as a complementary role for conversation in exploratory tasks. Because the study is observational, it does not show that assistant use caused better purchases.
The practical distinction is easy to recognise. “Find a room for these dates” specifies a searchable item. “Where could we stay so that one person can walk easily and the children still have things to do?” asks for help framing a choice. The latter is an illustrative question, not a quotation from the research dataset.
An assistant might help turn that broad request into a shortlist, after which the customer still needs dates, locations, prices and conditions. Returning to ordinary search need not mean the conversation failed. It may mean the user is completing a different part of the decision with a tool better suited to checking individual details.
What this means when planning a trip
For travellers, the sensible division is between exploration and confirmation. Conversation can be used to organise preferences, formulate questions and compare possible approaches. Before spending money, the decisive information should be checked against the actual booking offer: what is included, which dates apply and what happens if plans change.
A practical example is an assistant suggesting that a hotel suits a family. That description leaves several questions unanswered. Does the specific room accommodate everyone? Is breakfast included in the selected rate? Is access to an advertised facility restricted? A fluent overall recommendation should not substitute for the conditions attached to the particular purchase.
This is also why an explanation of trade-offs can be more useful than a single supposedly perfect answer. A traveller choosing between a central hotel and a larger property outside town needs to see what they gain and give up. An assistant that exposes those choices is easier to assess than one that simply declares a winner.
Readers can compare this with LiveAIWire’s coverage of AI-generated review summaries and purchase intentions. A summary changes how a choice is presented. That is separate from verifying whether every detail relevant to a particular customer is correct and complete.
Why retailers should be cautious with demographic labels
For businesses, the finding argues for testing assumptions about the audience. A design team that pictures only a young, technically confident early adopter may overlook customers who have a strong practical reason to use the feature. Conversely, a higher adoption rate in one older group does not justify treating all older customers as having identical preferences.
Age is a description, not an explanation by itself. Familiarity with the platform, frequency of travel, account history and the kinds of questions people want answered may all be relevant to understanding the pattern. An observational association cannot automatically disentangle those possibilities, even when the number of records is very large.
A useful product test would therefore ask whether customers can find the assistant, understand its role and return easily to conventional controls. It should examine the actual decision being attempted. A task such as locating an attraction and a task such as changing an existing reservation may call for different forms of assistance.
Success should also be defined before it is measured. More conversations could mean greater usefulness, but could also mean customers need repeated clarification. Fewer clicks could indicate efficiency, or a loss of opportunities to compare. A good evaluation would connect interface activity to understood, accurate and satisfactory decisions rather than celebrate activity alone.
The industry is connecting conversation to checkout
The distinction between advice and action has become commercially significant. In January 2026, Google introduced the Universal Commerce Protocol, describing an open standard connecting agents, businesses and payment providers across discovery, purchasing and subsequent support. That announcement concerns transaction infrastructure; it does not independently explain the demographic pattern found on Ctrip.
LiveAIWire’s earlier article on AI in retail and automated checkout examines that broader direction. The Ctrip study adds a different perspective: infrastructure capable of completing a sale does not tell a retailer how customers will choose to use the conversation leading up to it.
For a consumer, permission to suggest should not be confused with permission to spend. A system that explains several options occupies one role. A system authorised to place an order occupies another. The transition should make the selected item, amount and applicable conditions clear enough for the buyer to recognise what they are agreeing to.
A UK government analysis of agentic AI and consumers similarly identifies user intent, transparency and accountability as central to consumer value. It discusses shopping and finance agents as an illustrative scenario, not evidence that every such service already works reliably. That distinction is essential when interpreting industry demonstrations.
A large sample still has boundaries
The sample size makes the recorded behaviour substantial, but does not make the population universal. Customers using a travel platform are selected by that activity. People who never use it, people buying different products and users in different countries may encounter different interfaces and incentives. A large dataset cannot remove a boundary built into its source.
The same caution applies to the word older. It is a comparison within the study, not evidence that pensioners dominate all AI shopping. Nor does the finding support a claim that younger consumers reject assistants. The strongest defensible statement is that the observed adoption pattern differed from the stereotype, within this particular service and definition.
There is also no reason to assume a first trial predicts permanent loyalty. An assistant can attract attention without becoming a routine part of shopping. A fuller assessment would examine repeat use, the tasks people return for, whether questions are resolved and whether the service remains useful after its novelty has passed.
The opportunity is a clearer decision
For retailers, the most productive implication is to design around uncertainty customers genuinely face. An assistant might be valuable because it helps somebody articulate a complicated requirement, not because it demonstrates the newest model. That is an editorial interpretation of the opportunity, rather than a sales effect measured in the Ctrip study.
For travellers, the corresponding benefit is agency: getting a better grasp of the available choices while retaining control over the purchase. The promising use of conversation is not necessarily to remove every step. It is to make the difficult steps more understandable, then leave reliable ways to confirm the details.
The headline surprise is that older users were more likely to try this assistant. The more durable insight is that AI adoption has a setting. A useful feature, placed where somebody already has a reason to act, may tell us more about the future of everyday AI than an abstract image of who counts as a technology enthusiast.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.