AI Travel Search and the Direct Booking Opening for Owners
The story is not that AI books your apartment. It is that AI now decides which checkout the guest lands on, and that decision has become contestable.
Sometime in the last eighteen months, using an AI assistant to plan a trip stopped being an experiment and became a default. Depending on which survey you trust, somewhere between 40 percent of global travellers, rising to roughly 60 percent among millennials and Gen Z, and 56 percent of travellers now use AI tools at some point in trip planning. The reflexive industry reaction was that ChatGPT would swallow the booking and put the online travel agencies out of business. That version is wrong, and the way it is wrong matters for anyone operating short-term rentals in Malta.
What actually happened to in-chat booking
OpenAI launched Instant Checkout in late 2025 and extended it in February 2026, and for a few months the e-commerce world treated in-conversation purchasing as inevitable. Then, specifically for complex categories like travel, it stalled. OpenAI shifted ChatGPT away from completing transactions and back toward discovery, moving checkout into third-party environments, a change that visibly relieved Booking Holdings and Expedia, whose shares rose on the news. The technical reality was blunt: scraping retailer and travel inventory was inadequate for real-time pricing and availability, so the model moved to rerouting the user to the merchant's own site to complete the purchase.
What emerged in its place is more useful to an independent operator than in-chat booking ever would have been. The settled 2026 pattern is that discovery happens inside the AI interface while checkout happens in a merchant-controlled environment. The AI forms the intent. The transaction lands somewhere you can own. In the redirect model, the shopper is guided from the assistant to the merchant's website or app, and the merchant, not the AI platform, owns the checkout.
For a property manager, that routing decision is the entire commercial question. The same high-intent guest can be delivered to a Booking.com listing or to your own booking engine. The difference between those two destinations is 15 to 25 percent of the booking value.
The bypass is real, but it is not automatic
AI-referred traffic is not just growing, it converts better than anything else in the stack. AI referral traffic to US retail sites grew 393 percent year over year in Q1 2026 and converted 42 percent better than non-AI traffic. Channel-level numbers are starker still: ChatGPT referral traffic converts at around 11.4 percent, ahead of direct traffic at 10.2 percent and paid search at 9.3 percent. The reason is structural. A guest who arrives from an assistant has already compared options and narrowed intent in conversation, so they land ready to book rather than ready to browse.
The counterweight is the part most operators miss. AI does not yet route to independents by default. In a multi-model analysis of European hotel prompts in spring 2026, Booking.com and Expedia dominated AI-generated recommendations, each surfacing in more than half of them. Worse for the independent: an audit of 131,000 properties across 30 countries found only about 16 percent of global hotel supply was visible in ChatGPT, Google's AI and Perplexity, with chain-affiliated properties far more likely to surface. The result is a two-tier market of properties an AI knows and properties it does not.
So the bypass favours whoever is legible to a machine, not whoever wants the commission saving. That is a problem an operator can actually solve.
The commission math that makes the work worthwhile
This only matters because the numbers are large and recurring. Most OTAs charge between 15 and 25 percent per booking, with Booking.com typically at 15 to 18 percent and effective rates reaching 25 to 30 percent once promotional tools are stacked. Airbnb's host-only model runs around 15.5 percent. Put concretely, on a Malta apartment booked for a week at 1,200 euro, an 18 percent commission is 216 euro that either leaves with the platform or stays with the owner, on every single stay.
At portfolio level the effect compounds. Net RevPAR, revenue after commission, improves measurably with every point of weighted commission removed from the channel mix, and shifting roughly 30 percent of bookings to direct moves the weighted average down hard. Professional operators generally target 30 to 50 percent direct bookings, and treat anything below 20 percent as dangerous exposure to OTA algorithm changes. AI routing is, for the first time, a discovery channel that can feed the direct side of that ratio without buying the traffic outright.
What makes a property visible to AI
The signals that decide whether an assistant names your property are not the signals that won Google rankings. AI systems weigh structured data, review volume and recency, consistency across platforms, third-party mentions, and clear descriptive content, and many of those are not controlled by the property alone. Assistants also pull heavily from sources operators tend to ignore. AI tools frequently rely on third-party material such as Reddit, Kayak and travel blogs, and niche travel blogs often outperform major guides, because sites written for machines rather than to sell a room get rewarded.
The technical floor is structured data. Product data that is not machine-readable through JSON-LD and schema markup is becoming invisible to AI agents, which makes structured data a commerce prerequisite rather than a nicety. On top of that sits answer-engine discipline: schema types like LodgingBusiness and FAQPage, self-contained dated answers, and identical name, address and phone data across your site, Google Business Profile, TripAdvisor and Booking, because inconsistent details quietly reduce an AI's confidence in your property. One Princeton study cited in the same work found dated statistics lifted AI citation rates by 31 percent and named expert quotes by 41 percent.
What CiaoStay is building for this
We treat AI legibility as an operations problem with a medium-term build, not a marketing campaign. Five pieces are already in motion.
First, measurement at the point that matters. The Meta pixel and server-side Conversions API now sit on the Kross booking engine itself, not only on the website, so a booking that originates from AI discovery or a direct campaign is actually recorded as revenue with its value and currency. Without conversion data at the engine, optimising for direct bookings is guesswork.
Second, structured, machine-readable listing data, so each property can be parsed by an assistant with confidence about location, price band and type rather than inferred from loose prose.
Third, answer-shaped editorial content, the kind you are reading, built on dated figures and named regulations, Legal Notice 92 of 2026 and the MTCA framework, precisely because those are the signals AI engines cite.
Fourth, consistency of property details across Booking, Airbnb, Google and our own site, removing the contradictions that erode machine confidence.
Fifth, a genuine direct price advantage passed through at the Kross checkout, so that when AI hands us a high-intent guest, the redirect actually converts instead of bouncing back to the platform the guest trusts.
Where this goes
The operators who win the AI-routing era will not be the ones with the cleverest captions. They will be the ones whose property data a machine can read and whose direct checkout converts the intent the assistant has already built. Malta is a useful place to watch this play out, because Legal Notice 92 is already separating professional operators from improvised ones on compliance. AI legibility is about to become a second filter on the same axis, and it will sort the market faster than the licensing did.
