The AI Agent Already Sells Marriott's Afternoon Tea. What Can It Sell of Yours?
In China, an AI agent can already sell you afternoon tea at a Marriott. Not the room — the tea. Priced, bookable, with a link.
It happens inside Fliggy, Alibaba's travel platform. Its free agent software has eight commands. Two belong to Marriott alone: one finds its hotels by brand or destination, the other sells its packages — afternoon tea, spa.
Every other hotel in the catalog is read through five filters: star rating, bed type, price, property type, distance to an attraction.
For an independent, that's the whole vocabulary. "Double room, 3 stars, 800 m from the old town."
The agent sells what it can read
- Chains load products one piece at a time. First rooms, then points, then restaurants and spas. Each new piece is one more thing an agent can sell.
- Volume isn't the same as being readable. IHG, Hilton and Hyatt each sold more than RMB 100 million on Fliggy in a single summer sale — and none has its own command.
- The West is on the same path. Google's AI Mode in the US takes bookings from ten partners, five OTAs and five chains. Packages aren't there yet. The direction is.
What doesn't exist in a format a machine can read doesn't exist for the agent. The breakfast on the terrace, the late check-out, the bikes, the tasting at the winery next door: for a guest, that's why she picks you. For an agent today, it's invisible.
No panic: the AI guest is still rare
- Under 1% of traffic. AI-referred visits to hotel sites quadrupled in a year and are still below 1% of visits (Bain, with Similarweb data).
- But they now convert like everyone else. A year ago, AI visitors on US travel sites converted 47% worse than other visitors. In July 2026 the gap was 1% (Adobe).
- The industry isn't ready either. Only 9% of travel companies say they're fully prepared for agents — down from 12% last year (Phocuswright).
So this isn't the moment to rebuild your strategy. It's the moment to tidy up your catalog — while the share is small and the conversion is already there.
Five things worth selling besides the room
Extras are already a fifth to a third of a hotel's revenue, depending on the property. The question is whether a machine can find them.
- Pick five. Breakfast, late check-out, parking, one experience, one package. The ones guests already ask for.
- Give each one a name, a price and conditions. What's included, when, for whom. Written once, like a product — not buried in a paragraph.
- Same words, same price, everywhere. Your website, your booking engine, your OTA listings. The OTAs are the door most agents walk through first — Booking.com is already building agents that answer property questions. A partner can only pass on what you give it.
- Measure once a quarter. Look for referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com in your analytics. Compare their conversion with Google organic.
The takeaway: the agent will sell what it can read — and for most independents today, that's a room and a price. This week, write down five things you sell besides the room, with a name, a price and conditions, and put them identical on every channel. That's how a 30-room hotel plays in the Champions League: the machine sells the extras, your team makes them worth the trip.
Diego
Hospitality Net — "In China, the AI agent can already sell Marriott's afternoon tea", Markus Busch, hospitality.today (October 9, 2026): Fliggy agent software with 8 commands, 2 of them Marriott-only (hotels and packages); five filters for all other hotels; IHG, Hilton and Hyatt each above RMB 100 million in Fliggy's 2024 summer sale; Google AI Mode with ten booking partners.
Hospitality Net — "The AI guest is still rare", Markus Busch, hospitality.today (October 7, 2026): Bain & Company report with Similarweb data (AI referrals to hotel sites quadrupled July 2025–July 2026, still under 1% of visits); Adobe (AI visitors' conversion gap 47% lower in July 2025, 1% lower in July 2026).
Hospitality Net — "At 21, WiT Singapore asks what's left for travel companies to own in the age of AI agents" (October 7, 2026): Phocuswright, 9% of travel companies fully prepared for agentic AI (12% a year earlier).
PhocusWire — "Booking.com develops internal platform to build modular AI agents" (October 5, 2026): AgentHub and a property Q&A agent.
Hospitality Net — "Hotel Upselling: Motivating Employees to Sell More or Serve Better?", EHL (October 8, 2026): ancillary income at roughly a fifth to a third of total hotel revenue, per industry benchmarks.
