Guests’ AI asks precise questions. The OTAs’ AI already answers. Does yours?

A guest planning a trip no longer asks their AI assistant whether you have a room. They ask for a quiet family room for late October, a pool the children can use, and the best deal for a returning guest.

Booking.com and Expedia already answer questions like that with their own AI assistants, built on the rates, photos and room descriptions your channel manager sends them.

So the question for your own touchpoints is simple: what can a guest get on your website that they can’t get on Booking.com? Not the content. The OTAs have that. What they don’t have is who the guest is, what you are prepared to offer them, and what is happening at your property right now. A data layer brings that together. An AI touchpoint on your website lets a guest’s AI agent use it.

This is part two of our series on AI agents in hotel booking. Part one showed how an agent books on your website. This part is about what it should find when it gets there.

Your channel manager gives OTAs your content, not your advantage

Your channel manager distributes rates, availability and often your content too: photos, room and unit descriptions and amenities, to Booking.com, Google Hotels, Hotels.com and every other channel it feeds. That is how it should be: a guest sees the same room everywhere.

It also means every channel can answer the same questions about your property that you can, and the largest ones now do it with AI. Between a third and two fifths of travellers in the UK, France and Germany use AI to plan trips, and close to nine in ten of them say it shaped an actual trip decision (Phocuswright, 2026). When a guest’s assistant compares your website with an OTA listing for the same hotel, the facts match. If your website has nothing more to say, the assistant has nothing to recommend you for.

The large chains have noticed. In October 2026, Wyndham told analysts it is building a direct booking path that AI agents can complete easily, because “then we capture that guest.” Morgan Stanley puts OTA commission at 10 to 20%, against around 5% for a brand’s own website (Skift). For a hotel with EUR 10 million in room revenue, moving 5% of bookings to an OTA costs around EUR 90,000 a year in commission, as we calculated in part one.

What never reaches the OTA is what your own touchpoints can sell

What your channel manager cannot distribute is what belongs only to your own touchpoints, and that is exactly what a guest’s AI would use if it could reach it.

• Who the guest is. Your CRM knows that this family has stayed twice, always in summer, and asked for a cot both times. OTAs run their own CRM. Your guests’ history with you is not in it.
• Your negotiation framework. Your sales team works within rules that exist nowhere else: when to offer an upgrade instead of a discount, how far to go for a returning guest or a long stay. Today those rules live with a few people at the front desk or on the phone.
• What is happening right now. The weather at the destination this week, whether the outdoor pool is still heated, which restaurant is open, what is on in town.
• What you only sell directly. Packages, promo codes, member rates, add-ons and experiences you keep for your own website and app.

Put those together and the same question can get a different answer. For example, a guest’s agent asks for a family stay in late October. The OTA’s assistant lists the rooms that fit. Your website could recognise a returning guest, point out that the indoor pool is the better choice with that week’s forecast, and offer a fourth night on the terms your team would offer by phone. Only one of those answers is a reason to book direct.

A data layer puts that in one place

That information sits in different systems: content in a headless CMS, guest history in the CRM, availability in the booking engine and PMS, offers in a personalisation engine, payments in a payment system, and your offer rules in people’s heads or a spreadsheet. A data layer brings those sources together, so every touchpoint above it reads from one place: your website, mobile app, call centre, AI agents and, later, the digital signage in your lobby.

For you, that means three things.

• Every answer is complete. It combines the guest’s history, live availability and context.
• Offers follow your rules. Every touchpoint, including AI, offers what your best person on the phone would offer and nothing more.
• Access is decided once. What AI may read, offer and change is set in one place, under the same consent rules as your CRM. Your AI touchpoint shares personal data only with an identified guest, and negotiates only within the rules you set.

A headless CMS is a fraction of this: it manages content, but it doesn’t know the guest or your offer rules. The AI connectors some software vendors now offer are a fraction too, because each one connects AI to a single system. The connector gets AI connected. The data layer decides what AI hears.

An AI touchpoint lets a guest’s agent use it

Part one showed how WebMCP gives your website defined actions that a guest’s AI agent can use: check availability, get an offer, book. On their own, those actions return roughly what an OTA returns. Connected to a data layer, they return what an OTA can’t.

That is what we call an AI touchpoint: an AI layer on your own website or app that talks to a guest’s AI agent through WebMCP and answers from your data layer. It recognises a returning guest when the guest chooses to identify themselves, builds an offer within the rules you set, and uses context the OTA doesn’t have. An agent can only recommend what it can find. This gives it something to find on your website that it won’t find anywhere else.

WebMCP itself is still a proposed standard: Google opened an origin trial in Chrome in 2026, and support in AI agents is still arriving. The data layer doesn’t wait for it. The same sources and rules serve your website, app and call centre today, and are ready when agents arrive in volume.

Where you start depends on what you already run

If each channel is updated by hand

This is typical for independent hotels, campsites and smaller groups. Your first step isn’t technology. Write down what only you know: what you or your front desk would offer a returning guest, what can be given without asking you, which packages you keep for direct guests. Then decide which system holds each of them.

Your website is the first touchpoint to read from them. It is the one a guest’s agent uses to book you directly, and with fewer systems and fewer people to agree, smaller setups can get there faster.

If you already run a headless CMS and integrations

This is typical for hotel chains, DMCs and DMOs. You have the content side of a data layer. Three gaps tend to remain.

• Guest data stops at the CRM. The CRM knows the guest, but the website shows everyone the same page.
• Offer rules live outside every system. The call centre’s framework isn’t something a touchpoint can read.
• Nothing governs AI. What an AI may read, offer and change hasn’t been decided yet.

Five questions show whether your own touchpoints have an edge

The off-season is the time to start the first touchpoint

Systems and processes are easiest to change when the property is quiet, and 2027 planning is happening now. Pick one area where your own touchpoints should win, such as returning guests or long stays, and run the assessment while the property is quiet.

How Infosit helps

For more than 20 years, Infosit has connected the systems hospitality groups run on: booking engines, property management, CRM, marketing platforms and payment gateways. We work with what you already have.

Step 1: Data layer assessment. We map which of your systems hold which data, and what an AI touchpoint could use from them for the area you choose. You get the scope, duration and price of the pilot. The assessment is paid, and its fee is credited against the pilot if you go ahead.

Step 2: Pilot with one AI touchpoint. We take one area where your own touchpoints should win, for example returning guests. We connect the sources it needs into the first part of your data layer, set the offer rules with you, and put an AI touchpoint on your website on top of it. We also set up access: what the touchpoint may read, offer and change, under your existing authentication. You get a working result, an evaluation and a roadmap.

From there, we add sources and touchpoints area by area. If you run a headless CMS, it becomes the content source of your data layer. Either way, the connections are yours.

Frequently asked questions