Is Your Hotel Ready for AI Agents to Book It?

Imagine your next guest never opens your website. Their AI assistant searches, compares and books the hotel for them.

Joe Schnur believes this is coming, and soon. Joe is a collaborator of mine. He offers itinerary services to hotels in Florida, invests in travel startups and uses several personal AI assistants himself. He suggested this topic, and his ideas shape this whole article.

First, a short definition. An AI agent is an AI assistant that does tasks for you. It can search, compare and book. It does not only answer questions.

I build hotel websites for a living, so Joe’s idea makes me think about my own work. I did not want to just repeat it. So first I explain Joe’s view in his own words. Then I test each part with research and with my own tests. When something is only a forecast, I say so.

Joe’s Big Idea: The Guest’s AI Agent Books the Room

In his first message to me, Joe described the change like this:

Your next customer may never visit your website, never see your ad, never enter a credit-card number and never click a Buy button.

Their AI agent may discover your business, evaluate your offering, negotiate the transaction, authorize the purchase and transfer digital dollars directly to you.

He shows the change as two chains:

Today

1. Human →
2. Search →
3. Website →
4. Checkout →
5. Credit Card

becomes

Joe's picture

1. Agent →
2. Agent →
3. Money

In simple words: today, a person searches, opens a website and types a card number at checkout. In Joe’s picture, the guest’s agent talks to the hotel’s agent. The payment happens between them.

Later, Joe sent me a second note. He saw this topic spreading fast. He pointed to something that already works: Tesla owners can ask Grok, the AI assistant in the car, to order a coffee at Starbucks without using their hands. Then he took the next step, to hotels:

Very soon, users will be able to ask “find me a 4 or 5 star hotel, with recent strong reviews, king bed and room service for tonight. Only the hotels that can communicate with that AI agent will have a shot at that booking. The market will move VERY fast in this direction. Vacation rentals will move slower but the ones that move fast will get bookings.

Why does he think so? In his latest reply, he wrote that “human based UI and controls are going away,” replaced “with personal assistants and automation.” He also wrote that “no one really likes to search or use websites.” (UI means user interface: the screens and buttons that people use.)

What Joe Expects: Three Phases

Joe does not say everything will change at once. He talks about phases. The first phase is happening now. The later phases are his expectation, not facts. He saw one thing himself: he asked Grok in his Tesla about booking hotels and vacation rentals, and it said “it is coming soon.”

Phase 1, now. Agents use the websites and platforms that already exist. They pay with normal credit cards. In his words:

the agents can use existing platform UI (phase 1). agent based transaction processes will happen next - agent to agent economy - it will move fast - easy, safe, fast, improved outcomes.

Phase 2, next. Agents talk to other agents. Joe thinks the hotels that are ready for this will win:

agents want to talk with agents. they move to the path of least resistance. so businesses that have an agent that knows everything about the business and can talk with booking agents will have a leg up.

(“Have a leg up” means have an advantage. “Path of least resistance” means the easiest way.)

Phase 2 or 3, later. The middlemen lose their role. A middleman is a company between the hotel and the guest. Booking.com and Expedia are examples. They are called online travel agencies, or OTAs. Joe’s view:

a middleman exists because the independent business can’t move fast enough. but agents will make this easier. I think agents allow for more direct booking and take out the middlemen and their costs - but that will be phase 2 or 3.

Joe also expects people to start using agents very quickly. He names Meta Muse, Instinct and Grok Bot as growing fast. He says Instinct “can’t keep up.” His reason: “americans love the easy button.”

Joe’s Advice to Hotel Owners

Joe’s advice is simple, and it may be the most useful part. He wrote:

they should start using agents, watch videos, talk to others - don’t resist and put their head in the sand.

(“Put their head in the sand” means ignore a problem and hope it goes away.) He adds that owners who need help should ask an expert. At the end of this article, you can ask for an introduction to Joe.

He also expects different speeds. Most vacation rentals, he says, are run by one owner and not by a business, “so they move slow.” Property managers already use software systems, so they “will move faster.” He told me he must make his own rental website ready for agents too.

Joe is open about what he knows and what he is guessing. He writes “I think” and “I’d be guessing” when he is not sure. Now I check his ideas one by one. For each idea, I show the evidence and then what it means for your hotel.

Is Joe Right That Phase 1 Is Already Here?

Yes. Agents already work through existing platforms, as Joe says. The traveler is still in control.

On August 27, 2026, Google started to roll out hotel booking inside its AI Mode. A traveler can compare rooms, read the cancellation rules and pay with Google Pay in one chat. Ten partners were live at the start: Booking.com, Choice Hotels, Expedia, Hilton, Hotels.com, IHG, Marriott, Priceline, Trip.com and Wyndham.

The hotel or the booking site still takes the payment. It sends the confirmation and gives support. And the traveler approves each step. So “never click a Buy button” is where this is going. It is not how it works today.

The coffee example is real too. In September 2026, Tesla added Grok Bot to its cars. Owners can ask it to order from Starbucks while they drive. Trade and owner sites say it also works with DoorDash, Uber Eats and OpenTable. But ordering needs a subscription of $300 a month, so it is for early users only. It still shows the pattern: one spoken request, and only the businesses the agent can reach get the sale.

ChatGPT also has travel apps from Expedia and Booking.com. Trade news says the traveler sees live prices and availability in the chat, but the booking finishes on the partner’s own site. Sabre, a large travel technology company, also launched APIs for AI agents, so they can search, book and manage trips through its systems. (An API is a way for one piece of software to talk to another.)

Joe also said that agents in phase 1 pay by credit card. The payment industry agrees. The big companies are building their rules around cards:

• Google published its Agent Payments Protocol on September 16, 2025, with dozens of payment and technology companies.
• Visa introduced its Trusted Agent Protocol on October 14, 2025. It helps shops tell a real agent from a harmful bot.
• Mastercard announced Agent Pay in April 2025. It uses digital tokens linked to one agent and one permission.
• Stripe and OpenAI released the Agentic Commerce Protocol. The business stays responsible for the sale.

One thing to notice: all ten of Google’s launch partners are chains or OTAs. No independent hotel is on the list. I come back to this below.

Is Joe Right That People Will Adopt Agents Fast?

Partly. The early numbers are strong, but they show interest, not trust.

Muse is Meta’s personal AI agent app. It launched on September 8, 2026. TechCrunch and CNBC both reported that its first downloads were faster than ChatGPT’s first days on mobile. This supports Joe’s point about speed. But it counts downloads in the first weeks, not long-term use.

Instinct is a startup in San Francisco with a personal AI agent. News on September 28 and 29, 2026 said it raised $1 billion at a value of $10 billion. Its founder, Noah Shinn, said on a podcast that more than half of its payments are for travel. He also said it is close to $1 billion in yearly payments. These are his own claims, and nobody has checked them. I also could not check Joe’s statements about Grok Bot’s growth.

What do travelers say? SiteMinder, a hotel software company, ran a survey of almost 12,000 travelers in 14 countries (its Changing Traveller Report 2026). It found that 8 in 10 travelers want AI help when they book. That is almost four times last year’s number. But the top requests were about protection, not booking: 44% want price alerts, 39% want scam detection and 35% want help to watch their spending.

AltexSoft, a travel technology publisher, reports that only 2% to 8% of travelers would let an AI buy something without their approval. I could not trace this number to its original survey, so treat it as a rough signal.

My reading is this: wanting help is not the same as giving the AI the booking. The two surveys ask different questions, so I would not compare them directly.

For the longer term, IDC, a technology research company, predicts that by 2030, AI agents will make 30% of travel bookings. That is for all travel, not only hotels. And it is a prediction, not a measurement. Some news about Phocuswright, a travel research company, goes further. It says 61% of travel companies are testing or growing their use of AI agents, and that about 40% of consumers would let an assistant book flights and hotels. I could not open the original reports, so these numbers are not confirmed.

My verdict: Joe is right about the direction and about early interest. How fast guests will let an agent book a hotel alone is still open.

I Tested an AI Agent Myself

Joe says agents will use existing platforms in phase 1. To see what this means for a hotel, I tested Instinct on September 30, 2026. You can reach it by text, WhatsApp or phone call. It does tasks like booking travel, ordering groceries and canceling subscriptions. It is invite-only for now.

I used its WhatsApp agent. WhatsApp belongs to Meta, so the whole conversation ran on Meta’s messaging system. I searched for Andel’s Hotel Berlin (listed as Vienna House by Wyndham Andel’s Berlin), found it and went through the booking steps. I paid through Instinct’s vault. The vault keeps the traveler’s payment details so the agent can use them.

I stopped before the final confirmation, so I did not complete a booking. But the flow works, and it works with a normal stored card. This supports Joe’s point that phase 1 uses credit cards.

The most useful detail is what Instinct told me about how it books: directly on the hotel’s own website. This is Instinct’s own description. I did not see a finished booking to confirm it. Also, this hotel belongs to a chain, so its website may be better than a typical independent hotel’s website.

But if it is true, the hotel had no special connection to Instinct. The agent just used the hotel’s website. This leads to a clear conclusion: the hotel’s own website must be something an agent can use.

A review by another writer describes the same pattern. Instinct runs its own computer with a browser and saved logins, and it works through the traveler’s accounts. The review also says agents still get stopped by CAPTCHAs and two-step logins. I did not meet one. But I stopped before the end and tested only one chain hotel. So treat this as a lead, not a fact.

I also looked for a way for a hotel to connect to Instinct. I found no public program, partner list or sign-up page. The news I read does not say which systems Instinct uses to book rooms.

There are two more signals. They are only signals.

• Shinn is reported to be testing a standard that lets the assistants of different users plan trips together. This is the nearest real step to Joe’s Agent → Agent. But it connects one traveler’s agent to another traveler’s agent, not to a hotel.
• Skift reports that Shinn spoke about charging shops a fee for sending customers. He mentioned boutique hotels paying up to 30% per booking. This is only an idea, not a real program. If it happens, it would look like the OTA commissions that hotels already know.

What an AI Agent Needs to Know About Your Hotel

Take a request like Joe’s example: “a 4 or 5 star hotel, with recent strong reviews, king bed and room service for tonight.” An agent cannot answer it from a nice website. It needs facts that it can read and compare. It needs five kinds of facts:

• Does the hotel qualify? A star rating from a source it can trust. Recent review scores, with dates and the number of reviews, not only an average.
• Does the room fit? The bed type for each room type, not only in photos. Whether room service exists, and its hours, because “tonight” depends on the arrival time.
• Is it available, and what is the price? Live availability for tonight. The total price with taxes and fees, so the agent can compare you with other hotels.
• What is the guest agreeing to? Cancellation rules, deposit rules and check-in time, written clearly.
• Can the agent finish the booking? A booking path that ends online, with no phone call or email.

Now read this list as a hotel owner. The gaps are easy to see. The bed type is often only in a photo caption. Room service hours are in a PDF or on a sign near the elevator. The star rating is on one listing but not on another. Availability sits in a booking engine that the agent may not be able to read. When an agent cannot find a fact, it may guess, or it may skip your hotel.

In my experience, these facts are usually spread across your website, your property management system (PMS), your channel manager and several listings. A person can put them together. An agent needs them clear and the same everywhere.

This matches what Google asks for. Its Universal Commerce Protocol for hotels tells hotels to use their existing central reservation system (CRS) or property feeds. A feed is a list of your data that software can read. Google also says the hotel stays responsible for the sale and keeps the guest relationship. For now, the program has a waitlist. Sign-up and technical details are listed as coming soon.

Can an AI Agent Finish the Booking?

Joe describes agents that talk to the hotel’s own agent, and even negotiate. Today the reality is simpler. A guest’s agent needs something on your side that answers a question and takes a booking. It can be an agent. It can also be a normal system with a standard way to connect.

Today this means an availability check or a booking request, not a price negotiation. I found no independent hotel that negotiates with a guest’s agent. If someone says they do, ask to see it.

What is real is the connection layer. Sabre’s APIs for agents and Google’s hotel standard are both ways for an agent to ask a hotel system a question and get an answer that software can read. The hotel that answers well gets considered.

A Way In for Independent Hotels: Your Channel Manager

Joe says the hotels that move fast will get the bookings. But how can a small hotel connect without building its own agent? One answer is the system that many independent hotels already use to sell rooms: the channel manager. It sends your rooms and prices to many booking sites at the same time.

SiteMinder is a channel manager. It says it connects more than 56,000 hotels, and it now offers AI distribution. I am a SiteMinder partner, so please keep that in mind. Everything below comes from SiteMinder’s own AI Distribution page. These are the company’s claims, and I have not tested them.

• Channels Plus MCP. SiteMinder says AI travel platforms can take bookings through the same program as normal OTAs, with one connection. (MCP is a standard way for AI tools to connect to other software.) For hotels already on Channels Plus, it says no extra setup is needed. The hotel pays commission only on completed stays.
• Direct Booking MCP. For Demand Plus customers, SiteMinder says the hotel can appear inside ChatGPT, Gemini and Claude, through an AI partner’s app. It shows live prices and availability, with a link to the hotel’s own booking engine. The guest then books directly with the hotel.
• DirectBooker. SiteMinder says it is working with this AI-focused booking platform to show direct prices across ChatGPT, Claude and Gemini. It says this is still coming.

AltexSoft explains the logic: it is much easier to connect an AI platform to one big system than to thousands of hotels one by one. If this is true, many independent hotels will reach agents through the software companies they already use. They will not need their own custom connection.

Three warnings. First, SiteMinder’s page says more AI platforms are “coming soon,” and it does not say how many bookings have come through. Second, its “8 in 10 travellers” number comes from its own survey, which I have not seen. Third, other channel managers may offer something similar, or nothing yet. Ask yours.

Guesty, a property management system for vacation rentals, also has an official MCP server. This fits Joe’s point that property managers will move faster. But it is a beta and read-only. An assistant can look up data, but it cannot create or change bookings.

Will Agents Remove the Middlemen?

This is Joe’s boldest claim, and people disagree about it the most. He thinks agents will allow “more direct booking” and “take out the middlemen and their costs,” in phase 2 or 3. He says “I think,” so it is an opinion.

First, look at how hotels lost control the last time. Guests started to search on Booking.com and Expedia, and hotels paid commission to be found. In SiteMinder’s 2026 survey, 26% of travelers now start their hotel search on an OTA. Google and other search engines are lower, at 21%. But 18% of the people who start on an OTA still book directly with the hotel.

Now look again at Google’s ten launch partners. Every one is a chain or an OTA. This is my own analysis, not proof of a trend. Maybe big partners with direct connections simply launch first. But it is the same pattern: the companies that already control the data get the first chance with the new middleman.

AltexSoft has a more careful view. It says agents will change the balance in hotel distribution, but OTAs will not disappear. It quotes industry experts who say OTAs “may lose their dominance as the primary interface, but they will remain important as inventory providers, trust layers, and transactional intermediaries.” Some trade writers also warn that the big AI companies could take over the guest relationship and the booking data, with more power than OTAs have today. In the sources I saw, this is an opinion, not something that has happened.

I think these views can all be true, in different phases. In phase 1, chains and OTAs come first, as the launch list shows. Joe’s later phases can happen if independent hotels are ready. Google’s hotel program uses property feeds and a CRS, the hotel keeps the guest relationship, and the channel manager route is open.

A hotel that is ready may not need a middleman. A hotel that is not ready will find that an OTA or a chain gladly answers the agent’s question for it.

Can Agents Already Pay in Crypto?

This is the part of Joe’s picture with the biggest gap between talk and proof, so I checked it carefully. Joe says stablecoin payments “are growing rapidly,” mostly behind the scenes. He describes the model as “settle in dollars but transact in crypto,” and says it is “much safer and lower cost.” (A stablecoin is digital money that keeps the same value as the dollar. USDC is a well-known one.)

What I found supports the technology, with limits:

• A hotel booking example exists. On June 5, 2026, Travala, a travel booking platform, launched a tool that lets AI agents search, reserve and pay for hotels with USDC on the Base network. It uses MCP and was live in Claude Desktop at the start. Travala says it covers more than 2.2 million hotels, including big chains. The traveler must still approve the final payment, so it is not fully automatic. These are Travala’s claims as reported by Cointelegraph. I have not tested the tool.
• Agents already pay with stablecoins. A report by Visa and Artemis, from July 14, 2026, uses live payment data from the blockchain. It shows AI agents making real payments with x402, a standard from Coinbase that lets software pay for online services in stablecoins. The payments are very small. Visa’s conclusion is a mix: cards for bigger consumer purchases, and stablecoins for tiny payments between machines.
• Not shown yet. I found no independent hotel or vacation rental that gets paid in crypto through an agent. Joe expects that card companies and Stripe will launch their own stablecoins, and that their “3% fees are at risk.” I could not verify this, so treat it as his expectation.

So crypto payment for hotel bookings is no longer only a forecast. The technology works for a hotel booking today. The open question is whether it becomes a common way to pay for travel. A hotel stay is a bigger purchase, and Visa’s split suggests that this kind of payment stays with cards for now. That is my reading, not a statement from Visa about hotels.

I asked Joe if he has seen an agent pay for a booking in crypto. He gave a fair answer: “phase 1 agents are using credit cards.” Both can be true. Cards are the main way to pay in phase 1, and stablecoins are an option that has started.

What to Do Now: Nine Steps for Your Hotel

Joe says “don’t resist.” The research says the same. Most of the work is clean information and a booking path that software can finish. You do not need new software to start. These steps come from my own reading of the sources above. They are my advice, not rules from any platform.

1. Try an agent yourself. Ask an AI assistant to find and compare hotels in your market. Then ask what it needs to know to book yours. It costs nothing. Write down what it gets wrong about your hotel. That is your to-do list.
2. Put your facts in one place. Room types, prices, rules, amenities and house rules must be the same on your website, your booking engine and every listing. If two sources disagree, an agent cannot know which one is right.
3. Make your website readable by software. Put your rooms, prices, rules and location on the page as structured data, not only in photos and paragraphs. (Structured data is code that labels facts, such as price and bed type, so software can read them.) Ask your web developer to check it. You do not need a special file for AI. On June 2, 2026, John Mueller of Google called the llms.txt file “purely speculative for now.” Google Search Central also says you do not need special AI files to appear in AI search.
4. Do not block agents. Mueller said this is the biggest problem for most sites. Check that your firewall, bot protection and robots.txt do not turn away real AI agents. Also check that a booking does not meet a CAPTCHA before it can finish. You and your developer must decide how to block harmful bots but welcome real agents.
5. Ask your software providers direct questions. Ask your PMS, channel manager and booking engine if they support Google’s hotel program and agent access. They are at very different stages. Ask for a clear answer and write down who gave it.
6. Keep a booking path that ends online. If a booking needs an email or a phone call to finish, an agent cannot finish it. Show the cancellation rules clearly and confirm right away.
7. Check your payment setup. Ask your payment provider if it supports payments started by agents through the card networks. Google, Visa, Mastercard, and Stripe with OpenAI are all working on this.
8. Join the waitlist. Google’s hotel program is open for sign-up while the technical details are still being written. Signing up is free.
9. Measure it. Look at your website analytics for visits from AI assistants. If you cannot see them today, fix that first.

Here is what to skip for now. Do not build a payment plan around stablecoins yet, because hotels have only just started. Do not pay anyone who promises a guaranteed place in AI results. There is no evidence for that yet.

What We Still Do Not Know

Joe and I agree on the direction. We do not know the speed. I don’t know how fast guests will let an agent book their hotel. A hotel stay needs trust, and the current launches keep the traveler in the loop for a good reason. I also don’t know if independent hotels will get access on fair terms, or if crypto payment will matter for hotels at all.

Three signals would turn Joe’s forecasts into facts:

• An independent hotel appears among Google’s booking partners.
• Google opens its hotel program to all hotels.
• An independent hotel is paid in digital money by an agent, and the traveler does not approve each payment.

None of these has happened yet. I will update this article when one does.