Agoda Structures 700M Hotel Images and Reviews With AI: An OTA Shores Up the Data and Transaction Layers of AI Commerce
Booking Holdings-owned Agoda is using AI to pair 700M+ hotel images with guest reviews in 40+ languages and has bundled hotels, flights, and activities into a single checkout. We read the rollout as an OTA building machine-readable product data and a firmer transaction layer for AI commerce.
Key Takeaways
- Agoda, the Booking Holdings-owned OTA, launched an AI feature that links 700M+ hotel images with guest reviews across 40+ languages by topic, then followed with real-time flight alerts and a single checkout covering hotels, flights, and activities
- Converting unstructured images and reviews into a shared topic taxonomy is machine-readable product data work, the foundation AI needs to understand and present products, and a sign of an OTA shoring up its transaction layer as discovery shifts to AI assistants
- Structuring reviews and photos into units of extractable meaning, and keeping transactions and post-purchase service in your own hands even as discovery moves to external AI, are implementation questions retail and e-commerce operators share
Agoda's AI Rollout Is More Than a Display Upgrade
Agoda introduces AI-powered booking features that connect hotel reviews with images, making travel planning easier and boosting app engagement worldwide.
www.asiabusinessoutlook.comAgoda, the Singapore-headquartered online travel agency (OTA) owned by Booking Holdings, has been shipping a series of AI features across its app and website. On May 6, 2026, the company announced a feature that automatically pairs hotel images with the guest reviews that relate to them, processing more than 700 million images and millions of reviews in over 40 languages. In late June, a further set of features followed, including real-time flight alerts and a single checkout.
At first glance this looks like steady usability work on a travel app. This article reads it instead as groundwork for AI commerce, meaning selling in which generative AI and agents handle the journey from discovery to purchase, on both the data side and the transaction side. A company holding more than 6 million properties, 130,000 flight routes, and 300,000 activities has started reshaping its reviews and photos into a form AI can work with, and bundling multi-product payments into one transaction. Both moves raise questions that apply to any business with products, inventory, and reviews, whether it sells hotel rooms or physical goods.
Pairing Images With Reviews Is Product Data Structuring
The new feature displays relevant hotel photos alongside actual guest comments, organized by topics such as room cleanliness, pools, or breakfast. Each topic carries up to 15 images and review quotes shown in the traveler's own language, plus a sentiment breakdown, an automated read of how positive, negative, or neutral the feedback runs. The cross-checking of photos against reality that used to mean bouncing between the gallery and the review section now fits on a single screen.
Choosing a hotel always carries the worry that an attractive photo may not match reality. The Traveler reports that the initial rollout emphasized city hotels in Asia, where review volume is heavy and competition fierce. The higher the cost of verifying a product before purchase, the more a feature like this pays off.
The machinery behind it is documented in an engineering write-up covered by InfoQ. Classification models convert images into semantic labels such as pool or breakfast area, normalized into a canonical topic taxonomy. Reviews pass through natural language processing pipelines that extract representative snippets and sentiment signals, aligned to the same taxonomy. Two very different kinds of data, photos and text, become workable through a shared layer of meaning.
Generalize what is happening here and it amounts to making unstructured data machine-readable. Reviews and photos are rich, user-generated sources of information, but in raw form they are lumps that neither AI nor search handles well. Agoda re-cut them into units of meaning called topics and added a normalization layer that keeps semantics consistent across more than 40 languages. CTO Idan Zalzberg has emphasized the value of connecting what travelers see with what other guests experienced.
For now, the feature has been announced as a display improvement for human users, and Agoda has not stated any further use. Still, data whose meaning can be extracted topic by topic is the same foundation needed to design conversational search, recommendations, or product feeds to external AI channels. This is exactly why product data readiness is treated as the starting point for agentic commerce, the stage of commerce in which AI agents run comparison and transactions on the user's behalf.
What the Single Checkout Says About the Transaction Layer
Following the image-review pairing, Agoda announced a set of features in late June built around real-time updates and trip management. Line them up and it becomes clear that the company is polishing not the entry point of discovery, but the transaction and everything after it.
| Feature | What it does |
|---|---|
| Image-review pairing | AI matches hotel photos with relevant guest reviews by topic, with up to 15 images and a sentiment breakdown |
| Real-time flight alerts | Sends iOS and Android users an average of 2-3 notifications per booking covering flight status, gates, and baggage carousels |
| Gallery view (iOS) | A photo-forward browsing format for hotel search results |
| Single checkout | Book hotels, flights, and activities in one transaction, managed together in 'My Trips' |
| Expanded chat languages | Adds five languages for flight-related queries, including Chinese (Simplified/Traditional), Thai, and Indonesian |
The most telling of these is the single checkout. Hotels, flights, and activities, three separate products, can be booked in one payment and managed together in My Trips. The company invested heavily in expanding its flight business in 2024 and has been pushing to make the entire itinerary bookable on one platform. Bundling multiple bookings into a single transaction is an extension of that strategy.
A design that treats the whole itinerary as one transaction has the same shape as a future in which AI agents arrange multiple products at once. It is, however, a bundling of the booking experience inside Agoda's own app. There is no confirmed announcement that Agoda supports payment delegation to external agents, or connectivity standards such as MCP, a specification linking AI to external systems, or UCP, a proposed protocol for commerce across platforms.
Post-purchase experience is getting attention too. Flight alerts deliver status, gate, and baggage carousel information in an average of two to three notifications per booking. Chief Product Officer Ittai Chorev explains the intent this way.
Travel is always a little stressful, so having the right information at the right time is both useful and reduces stress.
Bundle the payment, then keep the post-purchase touchpoint alive through notifications. Unglamorous as they look, these two investments thicken exactly the areas external AI struggles to replace: transaction reliability and after-sales care.
Where OTAs Dig In as Discovery Moves to AI
Agoda's investments are inseparable from a structural shift in which travel discovery moves to AI assistants. The turning point came in March 2026. According to Skift, OpenAI moved ChatGPT's center of gravity from completing transactions to product discovery, delegating booking execution to third-party apps such as Expedia and Booking.com. Prices that move by the minute, cancellation policies that differ by rate type, and payments that cross currencies proved to be walls, as analysis of the shift points out.
Parent company Booking Holdings has been explicit about its posture. On its April 2026 earnings call, CEO Glenn Fogel said the company is working with external partners including OpenAI and Google so its supply is discoverable wherever travelers begin their journey. Spread discovery across external AI, close the transaction in-house: that is the division of roles. Assume this separation of discovery and checkout, and what Agoda is polishing is clearly the transaction side of the field: real-time inventory and pricing, dependable payment, and post-purchase service.
External AI assistants are not the only party in this tug-of-war. Suppliers are moving too: hotel chains have begun building conversational discovery into their own channels, as Marriott did with Ask Bonvoy. For an intermediary like an OTA, structured reviews and images and the quality of the transaction experience are also the assets that keep it chosen over direct booking.
Raise the experience on your own app while preparing to be found by external AI. That two-front posture is one instance of the design decision over how to divide the open garden and the walled garden. Losing the entry point of discovery and losing control of the transaction are not the same thing. A base that can absorb the transaction wherever discovery happens is becoming the OTA's line of defense.
Implementation Questions for Retail and E-commerce Operators
Bring Agoda's case home and the starting point is how you treat reviews and images. On many e-commerce sites, customer comments and product photos live in separate places, leaving pre-purchase verification to the shopper's patience. Linking them by topic or attribute and converting them into structured data with extractable meaning feeds not just your site's conversion rate, but whether AI can understand and present your products accurately.
Multilingual normalization carries the same weight. You do not get to choose the language or phrasing of questions arriving through external AI channels. As Agoda did by aligning reviews in more than 40 languages to one topic taxonomy, a data layer that keeps meaning consistent determines the accuracy of the answers.
Just as important is a design that keeps transactions and post-purchase service in your own hands even as the entry point of discovery moves outside. Bundle multi-product payments, and hold on to the post-purchase touchpoint through notifications and unified management. This territory is hard for external AI to replace, and it is where purchase data and customer relationships accumulate in your favor.
Conclusion
Agoda's AI rollout looks like a display upgrade built on pairing images with reviews, but in substance it is two kinds of groundwork. One is converting unstructured content into machine-readable data organized by topic. The other is bundling multi-product payments and post-purchase care into its own transaction layer. As discovery shifts to AI assistants, this is how an OTA digs into the transaction side of the field. For retailers and e-commerce operators holding products, inventory, and reviews, the question is the same: can you shape your content into a form AI can work with, and build a design that never lets go of the transaction and the customer relationship, wherever discovery moves next.
