AI agents and hotel booking: is your website ready?

Oct 22, 2026

When a guest asks an AI assistant in their browser to book a stay, the assistant opens your website and tries to finish the booking: check availability, compare rates, book.

Most hotel websites weren’t built for this. The agent misreads the page, makes mistakes or gives up, and books where it can finish the job. If that is Booking.com, the cost adds up fast. Take a hotel with EUR 10 million in room revenue: if just 5% of its bookings go to an online travel agency (OTA) this way, at an average 18% commission that is around EUR 90,000 a year.

How agents use your website today

A browser AI agent sees your website roughly the way a guest does. It loads the page, recognises buttons and fields, and tries to work out where to click. On a simple page, that works. In a booking flow with a calendar, room selection, packages and cancellation terms, every step is another chance to go wrong.

The guest never sees any of this. They only see the outcome: the agent reports that the dates are unavailable, that it couldn’t find a price, or it suggests another hotel.

Ten places where agents get stuck on hotel websites

Booking websites are designed for people who can see the page, understand context and click patiently. Tick each item that applies to your website:

Hotels and hotel groups

☐Booking engine on a different domain or subdomain. The agent has to carry dates and guest numbers across to another site and loses context along the way.

☐Group website without cross-property search. The guest asks for “anything available at any of your hotels”, and the website makes them pick a property first.

☐Landing page offers the booking engine doesn’t know about. The page says “stay 7 nights, pay for 6” or “15% early booking discount”, but the engine needs a promo code or never applies the discount.

☐Total price revealed only at the final step. Tourist tax, pet fees and parking are added at the end, so the agent can’t compare the real price.

☐Rate plans with terms hidden in tooltips and pop-ups. Non-refundable, flexible, half board: a person hovers to read the conditions, an agent never sees them.

☐Marketing room names. “Superior Family Comfort” doesn’t tell the agent whether the room has two separate beds for the children.

☐Bot protection and cookie banners. Protection that rightly blocks scrapers can’t tell them apart from an agent booking on a guest’s behalf.

Campsites and resorts

☐Pitch selection on an interactive map. A plus for the guest, a dead end for the agent.

☐Rules that change by unit type. Pets are allowed in the zone B mobile homes but not in the glamping tents. A person puts this together. An agent can easily miss it.

☐Extras in a separate step. Supplements for a seafront pitch, electricity or a second car come later, so the agent doesn’t know what the stay really costs.

If two or more of these sound familiar, your website works for guests but not for the agents booking on their behalf. The same issues are already costing you bookings from guests who book themselves. Fix the booking flow for guests and agents at the same time, and one project solves both.

Not sure where you stand? Find out where an AI agent is most likely to get stuck on your website. The questionnaire is free and you fill it in yourself.

Request the self-assessment questionnaire

WebMCP in plain terms

WebMCP lets your website tell an agent exactly which actions it offers and what each one needs. Instead of working from the page layout, the agent uses the actions you’ve defined.

For a hotel or resort, that usually means four actions:

Check availability: available units for the requested dates and number of guests.
Show offer: prices, packages and terms, exactly as you define them.
Book a stay: books through your system once the guest confirms.
Answer a question: parking, pets, check-in and other common questions.

You stay in control of two things. You set the sales rules (prices, terms, minimum stay) and the agent simply applies them. And every booking still goes through your own booking system, confirmed by the guest.

WebMCP is a draft standard developed in a W3C community group and edited by experts from Microsoft and Google. Chrome is testing it now. Other browsers haven’t committed yet. It’s early days, but the direction is clear: browsers are getting ready for agents that work with websites directly rather than reading screenshots.

This is where hotels start to split into two groups. On a website that offers actions, the agent knows what it can do: check availability, show the offer, book. On a website that doesn’t, the agent is left to work out the page from its layout, and it makes mistakes or goes elsewhere. As more guests hand their bookings to agents, that difference shows up in your direct bookings.

Comparison: a website the AI agent has to work out on its own versus a website that offers clear actions

Why this matters for hotels now

Few online purchases involve as many steps and rules as booking a stay. That’s why booking is where agents get stuck most often, and where hotels have the most to lose.

Guests have already moved. Between a third and two fifths of travellers from the UK, France and Germany use AI to plan trips, and 88% of German travellers who use it say it influenced a real decision (Phocuswright, 2026).

The big players are already there. Google has started rolling out hotel booking inside AI Mode, with OTAs and large chains as the first partners. Booking.com and Expedia already have apps in ChatGPT. When an agent decides where to book, it picks the place where it can finish the job.

OTAs are building for agents faster than hotels are. Every booking an agent sends their way carries a typical commission of 15 to 25%. At an average of 18%, a hotel with EUR 10 million in room revenue that loses just 5% of its bookings this way pays around EUR 90,000 a year. The maths works in your favour too. An agent-ready website turns this new channel into direct bookings, and a guest who has booked direct once is easier to win back without an intermediary.

Why now?

1. The rules are being set now

WebMCP is being drafted and tested in Chrome right now. In September 2026, Google began showing direct providers separately from intermediaries in hotel search in Europe, and says AI answers will follow. Hotels that prepare now are in from the start instead of playing catch-up with OTAs later.

2. Guest habits are still forming

Guests who make their first AI bookings through an intermediary will probably stick with it. Every season you wait means more guests to win back from OTAs later.

3. Preparation takes time, and winter is the time to do it

An audit, booking engine integration and testing don’t happen overnight, and a booking website is easiest to change off-season. 2027 season planning is happening now.

What this is not

Not another chatbot. A chatbot answers questions on your website. An agent-ready website takes bookings from the assistants your guests already use.
Not a new website. WebMCP builds on the data in your existing website and booking system.
Not a replacement for your booking engine. The agent uses your engine, just through a clearer route.
Not handing sales over to AI. The agent does only what you allow, and the guest confirms every booking.

How to start

For more than 20 years, Infosit has built booking flows and booking engine integrations for hospitality groups, including integrations with booking.com, Sabre and Phobs. The self-assessment shows where you stand. The audit shows exactly what to fix and in what order.

Step 1: Self-assessment. Start by finding out where you stand. You fill in the questionnaire yourself, free of charge, and it shows where an agent is most likely to get stuck on your website.

Step 2: AI agent readiness audit. If you want to know exactly what to fix and in what order, the next step is the audit. We test your website with an AI agent, and you get a plan with a scope estimate in time for 2027 season planning.

Step 3: Implementation. We build AI agent actions (WebMCP) for your key booking flows, connect them to your existing booking system and sales rules, and test them with AI agents in browsers that support them.

Ready for step 1? Request the self-assessment questionnaire and we’ll email it to you.

AI agent readiness audit

What you get from the audit

A clear view of how an AI agent sees your website today. We run through search, offers, booking and enquiries and show where the agent gets stuck, makes mistakes or gives up.
A list of the actions your website should offer. Specific to your hotel: which actions, in what order, and what each one needs from your booking system.
A rollout plan with a scope estimate. Steps, priorities, and what’s needed from you and from your booking engine vendor.
A one-page summary for management. To support the internal decision and budget planning.

What you can do after the audit

Decide on facts. You’ll know exactly where you’re losing agent-driven bookings and how much work it takes to fix that.
Give your vendor clear requirements. With the list of actions, you know exactly what to ask your booking engine vendor for, instead of a vague “do you support AI?”
Start implementation without guesswork. The scope is already defined, so implementation can start straight away. Hotel groups can begin with a pilot at one property.

Is your website ready for the 2027 season?

Find out where an AI agent gets stuck on your website.