Key Takeaways
- A Tom's Guide writer used the Viator integrations in both ChatGPT and Claude to plan a seven-day Kyoto trip, then compared the quality and presentation of each AI's suggestions.
- Both AIs surfaced real Viator tours with star ratings, review counts, durations, and prices, but Claude was preferred for assembling a full itinerary and adding experiences beyond Viator's catalog.
- AI clearly delivered on discovery and recommendation, yet actual booking and payment still sit outside the chat — leaving OTAs and activity operators with the homework of making their inventory AI-readable.
A Writer Hands an Entire Kyoto Trip to AI
Mapping out my future week-long stay in Kyoto
www.tomsguide.comIn June 2026, a Tom's Guide writer ran an experiment, published here, connecting Viator to both ChatGPT and Claude and asking each to plan a seven-day solo trip to Kyoto. The brief was specific: a 3,500-dollar budget and interests spanning video games, food, wrestling, vinyl shopping, anime, and photography. Work that once meant trawling travel blogs was handed wholesale to AI.
What this test reveals is not simply which chatbot is smarter. Because OpenAI and Anthropic each folded Viator into their ecosystems, agentic travel — where an AI conceives a trip and connects to an activity-booking platform to propose it — has reached a stage you can evaluate as a real, lived experience. The writer's account brings both its reach and its limits into sharp focus.
Viator appears in ChatGPT as an "app" and in Claude as a "Connector" built on MCP, the Model Context Protocol. How two AIs with different design philosophies handle the same underlying source says a lot about each company's agent strategy.
How ChatGPT and Claude Each Worked With Viator
The writer fed both AIs an identical prompt: act as an elite travel advisor and rank the most memorable Viator experiences matched to the stated interests and budget by value and likelihood of becoming a highlight. The gap between the two outputs for that single input is the heart of the experiment.
ChatGPT's Viator app integration pulled directly from Viator's live database, displaying options complete with star ratings, review counts, durations, and pricing. From a long list of cultural walks and street-food crawls, four picks rose to the top: a private Kyoto tour, a Nishiki Market brunch walk, a Gion geisha-district walking tour, and a Gion night foodie tour with nine-plus dishes and six sake tastings. Referencing real, live inventory on the spot is the basic form of agentic commerce.
Given the same prompt, Claude's Viator Connector surfaced cooking classes, sports lessons, theater shows, movie-location tours, and culinary tours — again with star ratings, review counts, durations, and per-person prices. To this point the two were roughly even. The difference showed up in presentation.
The decisive distinction was that Claude's Connector laid out the Kyoto experiences as a single, sequenced playbook. Claude also reached beyond Viator's catalog: a day trip to Osaka for the Super Potato retro-gaming store, a tour of Kyoto record shops, the Kyoto International Manga Museum, and a K1/kickboxing event — additions that leaned directly into the writer's hobbies. The writer credited both AIs with strong tour picks, but concluded that Claude's experience, which designed an itinerary and generated interest-aligned plans, was the one he preferred.
The Fault Line This Comparison Exposes: Discovery vs. Booking
The writer's verdict ends at "Claude wins," but read through an operator's lens, a more important structure emerges. What both AIs did beautifully was travel discovery and recommendation — not booking and payment.
Nowhere in the piece does the writer say he actually booked a tour. He only notes that he will keep Claude's advice in mind when the time comes to travel. In other words, the AI handled narrowing the field and designing the itinerary, while the final purchase stayed in human hands. That dividing line captures exactly where agentic commerce sits today.
Indeed, in March 2026 OpenAI moved its in-chat "Instant Checkout" out of the spotlight, shifting toward routing users to merchants' own apps and sites. Discover in chat, complete payment on the merchant's side. Skift reported that shares of Expedia and Booking Holdings rose on the news. The early fear that AI would seize the entire travel-booking funnel has, for now, receded.
The nature of travel as a product deepens this fault line. Hotels, tours, and transport carry volatile inventory, with dates, party sizes, and cancellation rules tangled together. It is harder than buying a single apparel item, and the trust bar for an AI to lock in a final confirmation is high. That is precisely why discovery and recommendation are where AI adds value today, while the booking funnel is bridged to the operator's own platform — the practical answer for the time being.
The flip side is that if AI does not discover you, you never reach the booking arena at all. The four tours that ranked highest for the writer's prompt were picked up precisely because they carried structured data such as star ratings and review counts. With AI-driven traffic reportedly up around 400 percent year over year across retail in a quarter — and 2,000 percent for some retailers — designing experiences to be findable by AI is becoming a precondition for acquisition.
What OTAs and Activity Operators Should Prepare For
Read from the standpoint of travel and booking businesses, the homework is clear. It distills to one question: is your inventory in a state that is easy for an AI agent to read and easy to choose?
Viator was embedded in both ChatGPT and Claude because it had structured its inventory, pricing, ratings, and durations and opened them in a form AI can reference. Tripadvisor's activities, under Viator's parent company, have also become available inside Claude, accelerating the move to hold a storefront directly inside AI platforms. The shift is from waiting for traffic to your own site toward delivering inventory to wherever the AI resides.
This is where metadata pays off: star ratings, review counts, durations, and cancellation policies. Every experience the AI selected in the writer's test came with structured data to support a decision. Inventory described only with images and prose, thin on machine-readable attributes, struggles to make the AI's comparison table. Technical investment to support MCP and the various agentic-commerce protocols sits on the same continuum.
At the same time, what leading retailers like Target guard fiercely is the direct relationship with the customer. Even when discovery and recommendation are delegated to AI, the booking data, loyalty, and repeat touchpoints stay in-house. The balance of exposure to outside AI, relationship kept inside applies just as squarely to OTAs and activity operators. Preparing to be chosen by AI and building the mechanisms to nurture the relationship afterward must be considered together.
Conclusion
This experiment, handing a Kyoto trip to ChatGPT and Claude, shows that agentic travel has reached a practical stage in discovery and recommendation. It also makes plain that booking and payment still rest with the operator, and that a division of labor between AI and operator is the working answer for now.
For travel and booking businesses, the dividing line is whether they can shape inventory into a form AI can read and deliver it to where AI gathers. The trend of AI becoming the front door to a trip has moved beyond experiment and is now in motion.
