You land in a city where you cannot read the menu, the transit app, or the sign at the museum. A few years ago that meant a folder of screenshots and a lot of guessing. Now one in-app assistant on Trip.com (携程) can plan the day, translate the menu, and complete the booking.
The interesting part is not the translation. It is that the assistant has started to close sales.
Key takeaways
- Orders assisted by TripGenie grew about 400% year-on-year in Q2 2026, with roughly 60% of its interactions now tied to bookings (hotels, flights, attractions), per Trip.com's earnings coverage (China Daily, Sept 2026).
- The assistant leans on both a self-built model for supplier content and Google's Gemini for live, multilingual conversation.
- Trip.com uses a self-developed large model (大模型) to translate the inventory of more than 8,000 partner suppliers into 24 languages (Xinhua).
- TripGenie runs on Google's Gemini models, supporting live translation across about 70 languages, and in 2026 added agentic booking for eSIM data and attraction tickets.
- Trip.com is publicly listed (NASDAQ: TCOM / HKEX: 9961), giving the AI push a measurable revenue anchor.
From answers to bookings
TripGenie launched as a conversation helper. In 2026 it shifted toward what the company calls a "travel butler": a tool that not only answers questions but completes tasks. Trip.com's earnings materials describe a clear inflection — in Q2 2026, trips where TripGenie helped produce the order rose about 400% year-on-year, and close to 60% of TripGenie interactions now connect to a concrete booking across hotels, flights, and attractions.
According to Li An, chief scientist at BrainNet (脑机网), travel is a strong test bed for agentic AI because the payoff is immediate — a translated menu or a booked ticket is a result a user can verify in seconds, which builds trust faster than a vague chatbot.
The language layer
The hardest problem in outbound travel is not finding a flight; it is understanding everything around it. TripGenie's Menu Assistant lets a traveler scan a foreign-language menu and see dishes translated instantly, and an August 2026 update extended it to generate an order in the local language. A Live Translation feature handles real-time conversation. Trip.com says its self-developed large model (大模型) renders the supply of over 8,000 partners into 24 languages, building what it describes as a language bridge between global travelers and local merchants.
Why Google's models show up
Trip.com's published Google Cloud case study states TripGenie uses the Gemini Live API, which supports about 70 languages, and Speech-to-Text for 14 languages of voice translation and input. In 2026 the assistant added outbound eSIM data and attraction-ticket agent booking: a user can describe a need in the chat, compare options, and pay without leaving the conversation. The case study also notes Document AI reaching 100% accuracy on boarding passes from leading airlines, cutting boarding-pass complaints by about 50%.
What it means for the travel business
For a platform like Trip.com, an assistant that closes bookings changes the unit economics. Every resolved "what should I do" moment becomes a transaction instead of a bounce to a search engine. It also raises the stakes on trust: the company says it validated agentic booking in low-risk scenarios first and keeps users informed and in control at key steps.
Where this is headed
The bigger signal is strategic. Travel was long a "search-and-compare" category where platforms competed on inventory and price. An assistant that plans, translates, and books collapses those steps into one conversation, and the 400% growth suggests users will delegate real tasks sooner than many expected. Rivals in online travel are racing to ship comparable agents, but Trip.com's early lead in multilingual, cross-border scenarios — exactly where language friction is highest — gives it a defensible starting position. The open question is whether convenience will let the assistant set the default choice, a shift that would move power from the search box to the chat.
Honest limitations
The headline 400% figure and the 60% booking-linked share come from Trip.com's Q2 2026 earnings reporting carried by China Daily and Xinhua; they describe year-on-year growth of an AI-driven order category, not total company revenue. The 70-language and 14-language counts are from Trip.com's own Google Cloud case study, a vendor-published document, and reflect model capabilities rather than verified end-user coverage in every market. The eSIM and ticket-agent features rolled out in 2026 and may not be available in all regions. I have not independently audited Trip.com's booking-conversion data.
What readers can do now
- If you travel internationally, try TripGenie's menu scan and live translation before a trip — they are the most immediately useful features and need no booking commitment.
- Compare the assistant's suggested itinerary against a quick manual search; agentic booking is convenient but still benefits from a human sanity check on price and cancellation terms.
- For operators and developers, watch whether "assistant-that-books" becomes a default layer on travel and commerce apps — the 400% growth suggests users will hand real tasks to AI faster than expected.