The shift at the entry point — guests asking AI where to stay — has already been covered in our earlier work. This article looks at the next stage: since summer 2026, a path where the booking itself is completed inside a conversation with AI has actually started to operate, albeit on a limited basis. On August 27, 2026, Google began offering US users a feature in Search’s AI Mode that lets them compare rooms, check cancellation policies and pay, all on a single screen. Around the same time, a major short-term rental platform began testing lower fees for guests that hosts bring in themselves, and on September 2, Radisson Hotel Group announced its hotel search app on ChatGPT in a Japanese-language press release. This article is neither about content readiness (see our earlier piece, Google UCP & ChatGPT: 3 Data Fixes Hotels Should Make Now) nor about cutting commissions on existing channels (our earlier analysis “Look at inventory allocation before going OTA-free”). It focuses narrowly on what operators and headquarters should decide when one more sales channel is added. The conclusion first: at this point, few mechanisms apply directly to Japanese properties, and the right move is not “deciding whether to sign up” but “getting ready to be able to sign up.”

What Was Announced — Three Facts, and How Far They Matter for Japanese Properties

Let us first lay out the facts briefly. The moves reported since summer 2026 tend to be lumped together as “bookings completed inside AI,” but they fall into three groups. The first is Google’s feature that completes payment within Search’s AI Mode — the only path that literally qualifies as “the booking is made inside AI.” The second is hotel chains’ apps on ChatGPT that show availability and rates, which keep the place where the booking is made on the hotel’s own website. The third is a short-term rental platform’s fee reduction; it is not directly related to AI, but it bears heavily on channel design in that “how fees are set changes with the referral route.”

Source: Compiled by HotelBank Editorial Team from company press releases, Skift (August 27 and 29, 2026), Travel Voice (March 6 and September 2, 2026) and PR TIMES (September 2, 2026)

The table makes one thing clear. The only path that can be called “the booking is made inside AI” is Google AI Mode in the US, and even there the merchant of record stays the hotel or the OTA. In other words, even when AI completes a booking, which channel the revenue is booked under in the hotel’s accounts depends on which business sits behind the AI. Radisson’s case is a “discovery path” that hands off to the hotel’s own website, an extension of the trend we described for the first half of 2026 (discovery moves first, booking returns to each company’s own path). The short-term rental platform’s test changes pricing within a channel, altering the rate when the referral originates with the host. No mechanism that applies directly to Japanese properties can be confirmed at this point, but the fact that all three moves point the same way — “the combination of fee and booking location changes depending on who brought the guest” — is enough to set preparation priorities.

Looking at Japan, similar moves can be seen at two levels. One is a major domestic OTA, which on April 30, 2026 added booking to its own AI agent: after comparing options in conversation with the AI, the booking is completed in the company’s app via a booking button. The other is AI guest-service products for lodging operators. “talkappi ChatAgent,” launched by Activalues Inc. on August 10, 2026, is designed to integrate with PMSs and booking engines and complete everything from inquiries to availability checks and booking within the chat (some functions are described as under development). In the former, the entry point becomes AI but the booking is still made at the OTA; in the latter, the booking is made on the property’s own side. That distinction is the subject of the next section.

“More Entry Points” and “More Places Where Bookings Are Made” Are Two Different Things

Placing these cases on a single grid sharpens the issue. With “the entry point where guests first learn about a property” on the horizontal axis and “where booking and payment are completed” on the vertical axis, the 2026 moves scatter across four quadrants — and it is the vertical axis that affects the design of commission, inventory and pricing.

Source: Compiled by HotelBank Editorial Team from company press releases

In the upper-right quadrant — AI entry point, booking made at the OTA — nothing changes in the property’s P&L. Commission and inventory distribution follow existing contracts; the only change is which screen the guest saw the options on. The lower-right quadrant, by contrast, is an extension of direct sales, but whether “some form of compensation goes to the AI side” has not yet been determined. Industry commentary also notes that, for bookings executed by AI agents, “nothing confirmed and verifiable exists yet” regarding commission structures, and both Google and OpenAI keep each company as the merchant of record while holding the conversation side themselves. One more channel becomes real only at the moment a new commission rate is written into the lower-right quadrant. What we see now is the step before that: “more entry points” and “reshuffled rates within existing channels.”

Three Variables That Move When a Channel Is Added — Commission, Inventory Distribution and Price Consistency

So when a rate is written into the lower-right quadrant, what should operators and headquarters calculate? Following the same structure we used in our earlier analysis of inventory allocation before going OTA-free — “annual revenue per room × channel share × commission rate” — we set out how the annual cost per room changes when the new channel’s rate is 0% (referral only, booked on own website), 6%, 10% or 15%. Rates are the estimated settled ADR; occupancy is the Japan Tourism Agency’s 2025 final figure.

Annual revenue per room = estimated settled ADR × actual occupancy × 365 days (rooms revenue only). Estimated settled ADR is the 12-month average from August 2025 to July 2026 of monthly medians across 47 prefectures, weighted by number of properties (N = business 7,368 / city 1,096 / resort 1,550 / ryokan 6,684 properties). The 10% new-channel share and the commission rates are assumptions, not actual contract terms.

Source: Compiled by HotelBank Editorial Team from MetroEngines Research / Japan Tourism Agency, Overnight Travel Statistics Survey, 2025 annual figures (final)

Taken alone, this table leads to the obvious conclusion that “the lower the new channel’s rate, the better,” but that leads to the wrong decision. What really matters when a channel is added is where those bookings moved from. Assuming 10% of annual revenue moves to the new channel, we set the prior channel two ways — “existing OTAs and similar sites (assumed 10% rate)” and “direct sales on the own website (online payment 3.6% + booking engine usage fee 1.22% = about 4.8%, based on the fee stack in our earlier analysis of a ¥13,600 hotel night netting ¥10,000)” — and the chart below shows the difference in annual cost for one business hotel room.

Source: Compiled by HotelBank Editorial Team from MetroEngines Research / Japan Tourism Agency, Overnight Travel Statistics Survey, 2025 annual figures (final) (estimates for one business hotel room, annual revenue ¥2,446,000, 10% via new channel)

If the bookings moved from existing OTAs and similar sites, a new channel at a 6% rate saves ¥9,800 a year, and referral only with booking on the own website saves ¥12,700 a year. But at the same 6%, bookings that moved from direct sales turn into an extra ¥2,900 a year. At a 15% rate, a shift from direct sales costs an extra ¥24,900 a year. Scaled to 100 rooms, the same decision — “add one channel at a 6% rate” — can mean either ¥980,000 a year in savings or ¥290,000 a year in extra cost. The economics of a new channel are determined not by how high or low its rate is but by what it cannibalizes. Our earlier work framed “the value of 1pt of direct sales as annual revenue per room × 1% × commission rate”; with an added channel, note that the sign of that formula flips depending on where the bookings came from.

The second variable is how inventory is distributed. Google AI Mode inserts a step in which the user reviews the room and terms and approves payment, while billing and booking confirmation are handled by the hotel or OTA acting as merchant. This means it is a prerequisite that the booking engine or channel manager connected to the AI can supply inventory and rates as fresh as those on existing OTAs. If inventory on the new channel is treated as “released only when left over,” the AI will show no availability at the moment it presents the property, defeating the purpose of adding an entry point. Conversely, opening all inventory unconditionally just because the channel books on the property’s own side will break alignment with existing channels on high-demand dates. As shown in our earlier work, public allotments sell out faster on stronger-demand dates, so it is realistic to give the new channel a demand-date allocation policy from the outset. The share of rooms listed on booking sites itself has been covered in a separate analysis and is outside the scope of this article (for the distribution of allocation rates by property type, region and size, see also Median 39.2% of Rooms Online — Allocation Across 12,941 Hotels).

The third is price consistency, so-called rate parity. In an environment where AI collects and lines up the prices of the same room from multiple merchants, the effort of switching screens to compare drops to zero. In our earlier measurement of 9,030 like-for-like pairs, the share where prices on two sites did not differ by even ¥1 was 71.0%, and price differences were the result of design rather than chance. In front of AI, designed and undesigned price differences alike are fully exposed. It is telling that the short-term rental platform’s test is reported to let “hosts choose whether to pass part of the reduction on to guests”: if rates differ by channel, it becomes one step harder to design prices so that the official channel is not at a disadvantage under identical conditions. Before adding channels, you need to decide which condition’s price will serve as the baseline and what you will differentiate on through added value.

Implications by Size — New Supply Averages 24.3 Rooms, While 100+ Room Properties Hold 65% of Rooms

To consider who is most affected by an added channel, we need to look at Japan’s supply structure. Among properties tracked by MetroEngines Research that are confirmed as listed on major booking sites with known room counts, new openings in 2026 totaled 1,706 properties and 41,396 rooms, averaging 24.3 rooms, as of October 10, 2026. Compared with 1,985 properties in 2024 (average 19.7 rooms) and 1,772 in 2025 (average 23.4 rooms), the number of properties is declining slightly while the size per property is gradually increasing. Note, however, that because this is based on listing confirmation, openings from October 2026 onward will increase as they get listed.

Source: MetroEngines Research & Consulting (based on OTA listing confirmation; 2026 openings with known room counts, N=1,706 properties / 41,396 rooms, as of October 10, 2026)

Split by size band, the picture becomes twofold. Properties with 29 rooms or fewer number 1,406 and make up 82.4% of properties, but only 5,786 rooms, or 14.0% of rooms. Conversely, properties with 100+ rooms account for 137 properties (8.0%) but 26,861 rooms (64.9%). Measured by “the number of businesses that must respond,” the main actors in channel addition are small properties; measured by “the number of rooms that move,” they are large properties. This duality tends to cause headquarters and on-site teams to talk past each other.

Source: MetroEngines Research & Consulting (based on OTA listing confirmation; properties with known room counts in each year. 2024 N=1,985 / 2025 N=1,772 / 2026 N=1,706. 2026 is as of October 10, and the most recent months onward may increase as more listings appear)

Property counts and average rooms are both based on properties with known room counts (1,599 of the 1,706 total in the top 9 types). “Typical sales channels” is the editorial team’s summary of general tendencies by type and does not describe individual properties.

Source: MetroEngines Research & Consulting (based on OTA listing confirmation, as of October 10, 2026)

By type, the largest group of new supply in 2026 is vacation rentals, at 823 properties averaging 1.7 rooms; together with guesthouses, cottages and machiya, small accommodation-only types total 1,054 properties, more than 60% of the total. Many of these have no booking engine or loyalty program of their own, and the platform listing page effectively serves as the “official website.” It is no coincidence that the short-term rental platform’s fee-reduction test is designed precisely for this segment. The less able a segment is to maintain its own sales channels, the more directly its economics are shaken by pricing changes — “which platform applies what rate to host-originated referrals”. On the other hand, many of the 210 business hotels (average 98.9 rooms) and 51 city hotels (average 112.3 rooms) have HQ booking platforms, and just as the hotel participants in Google AI Mode were all major chains that connected their own booking platforms as merchants, connecting to AI can be handled as a question of HQ technology investment. For the former, AI channels are something to “wait for the platform to connect”; for the latter, something to “decide whether to connect themselves.” The same news calls for actions with different subjects.

What You Can Prepare Now — Structured Property Data, Centralized Rate and Inventory Management, and Written Cancellation Policies

Since the timing in Japan is uncertain, what to invest in now is not signing up for a specific service but a state in which you can connect by the same procedure whichever channel arrives. Google AI Mode’s specification shows a flow in which, before AI completes a booking, the user checks three things: “room details,” “cancellation policy” and “payment amount.” Properties where these three are not structured are less likely to remain among the options even when the entry point opens. To the three stages covered in our earlier piece on data readiness — static content, ARI (availability, rates and inventory) and review consistency — this article adds the following from the perspective of channel addition.

Source: Compiled by HotelBank Editorial Team

What the four items have in common is that none is a new investment; all are about organizing existing information, and none will be wasted regardless of whether a specific AI service launches in Japan. Item 4 in particular is easily overlooked, but as the estimates in the previous section showed, the sign of a new channel’s P&L flips with the source of the shifted bookings. A property that cannot measure where bookings moved from cannot tell, after adding a channel, whether it gained or lost. For properties in the 24.3-room average size band, adding one “origin” column to the monthly booking ledger is enough.

When to Act — Closing with Decision Criteria

The numbers in the area covered here are not settled. Google’s feature is US-only with no published commission structure, the ChatGPT app stops at discovery, and the short-term rental platform’s fee reduction is a reported figure. It would not be appropriate at this stage to declare “sign up” or “wait and see.” Instead, we set out criteria: once which facts are confirmed, which decision to move to.

Source: Compiled by HotelBank Editorial Team

To sum up, three facts were confirmed in summer 2026: “a path where the booking is completed inside AI has started in the US,” “apps on ChatGPT handle discovery and send bookings back to the hotel’s own website” and “even within platforms, rates have started to vary with the origin of the referral.” None of these forces Japanese properties into an immediate sign-up decision. However, since all three point the same way — the combination of fee and booking location is determined by who brought the guest — there is real value in building, now, the ability to calculate the P&L of an added channel by “source of shifted bookings” rather than by “commission rate”. The same preparation works differently for each: for the small properties that make up 60% of new supply, it is a hedge against platform pricing changes; for large properties with HQ booking platforms, it is the input for deciding whether to connect themselves.

Frequently Asked Questions

Related Reading

• Google UCP & ChatGPT: 3 Data Fixes Hotels Should Make Now
• Median 39.2% of Rooms Online — Allocation Across 12,941 Hotels