Booking Holdings just proved something most travel marketers still refuse to hear: AI only pays when you own the checkout. Citations without conversion are vanity metrics. Own-assistant conversion is the real scoreboard.
On the Q3 2026 earnings call, CFO and EVP Ewout Steenbergen said users of Priceline’s Penny and Agoda’s AI Assistant converted at roughly twice the rate of non-users, with Agoda Assistant users booking about 20% higher average value. LLM referrals into Booking.com? Still significantly less than 1% of room nights. Same company. Same quarter. Two completely different stories.
Own the assistant, own the booking
PhocusWire’s Cathy Walsh covered the numbers cleanly. Booking did not disclose how many people actually use those assistants, or the absolute conversion rates. So treat the 2× claim as directional, not a blank cheque. Still: when the CFO puts “twice the conversion” and “+20% average booking value” on the same call, he is telling investors where the money sits.
Glenn Fogel put the strategy in one line: these experiences show the difference between recommending a trip and actually enabling someone to book it seamlessly. That is not a product slogan. It is a commercial hierarchy. Recommenders get credit. Merchants get margin.
Agoda’s AI Assistant is now live for mobile users in 21 markets after a September limited beta. OpenAI powers the conversation. Gemini plus Google Maps Grounding handles restaurants and points of interest across more than six million accommodations, plus flights and activities. In beta, users were more likely to book connected multi-component trips. That matters more than a pretty chat UI. Connected-trip transactions on Booking.com grew nearly 20%, still a low double-digit share of the mix, while flight tickets rose 19% to 21 million. Assistants that stitch the trip keep the transaction inside the house.
External LLMs are still a rounding error
Skift reports the other half of the story. Organic LLM referrals rose after the latest ChatGPT models improved citation rates. Paid OpenAI referrals are growing as Booking expands markets and tests. Daily AI search visitors for trip search on Booking.com jumped 10× in Q3 versus Q2. OpenTable × OpenAI seated 190,000 covers in September — among the fastest-growing channels, from a small base, with Muse and other horizontal agents bringing new customers.
All of that sounds huge until you put it next to Steenbergen’s line: LLM referrals are still well under 1% of Booking.com room nights. Laura Houldsworth, Booking.com’s APAC lead, told Skift that travellers hop Claude → ChatGPT → Gemini → Google, then come direct. An earlier 2026 Booking survey said 64% of APAC travellers use AI to plan. Planning is not booking. Referral share under 1% is the proof.
I have been saying this for months. When I wrote about Booking’s AI Answers on hotel listings, the point was the same: the brand that answers inside its own funnel keeps the click. When Expedia shipped thirty AI features while European travellers asked for proof, I argued feature theatre without trust is noise. Booking’s Q3 numbers sharpen both takes. Own-channel AI converts. Third-party citations barely move room nights.
What this means for travel marketers
Stop celebrating “we’re cited in ChatGPT” as if it were a conversion channel. Treat it as awareness until your analytics show otherwise. Then ask the harder questions:
- Who owns the last click? If the assistant can recommend but cannot complete payment and servicing, you are training demand for someone else’s cart.
- Are you measuring assistant conversion, or only LLM referral? Booking separates them. Most hotel and DMO dashboards still mush “AI traffic” into one vanity line.
- Is your offer agent-bookable, or just agent-mentionable? Flight MCPs and OTA assistants are racing to be the merchant under the chat. I covered that when Etraveli opened its flight engine to AI agents. Citations without a checkout path are free advertising for the platform that can finish the trip.
- Connected trips beat single-SKU chatbots. Agoda’s beta users booked more multi-component trips. If your AI only answers FAQ, you are not in the same game.
Regulation will not save the lazy
Fogel also flagged that European regulation may slow horizontal personal AI agents. Fine. That buys hotels and airlines time. It does not buy them a strategy. If Booking’s own assistants already convert at roughly 2× while external LLMs sit under 1% of room nights, the winners will be the organisations that put AI on the path to payment, not the ones waiting for ChatGPT to send them free demand.
So here is my take, bluntly: AI in travel is not one market. It is two. Own-assistant AI is a conversion product. External LLM referral is still a research side-effect. Booking just published the scoreboard. Most of the industry is still cheering the wrong match.
Where do you sit — citations, or checkout?
Want more takes like this on AI, search and travel marketing? Subscribe to my LinkedIn newsletter.