At HITEC this June, in a distribution session running twenty minutes over, a general manager from an independent property stood up during Q&A and asked the only question that mattered all week. “Is ChatGPT going to do to us what Booking.com did?”
The panel did what panels do. Someone said “partnership opportunity.” Someone said “it’s early.” Someone plugged a product. The GM sat down without an answer, and you could feel the room register that he hadn’t gotten one.
We have heard some version of that question from nearly every hotel group we have signed this year. It is the existential question of the agentic-AI moment, and almost nobody selling into this industry will answer it honestly, because the honest answer is uncomfortable for most of the people holding a microphone.
So here is the honest answer, at full length. It ends better than the panel’s silence implied. But not automatically.
The fear is rational because the exact thing feared has already happened once.
Start with the memory every hotelier carries. In the early 2000s, online travel agencies showed up offering something that looked free: incremental demand, a global storefront, no upfront cost. Hotels handed over inventory to fill shoulder nights. Twenty years later the storefront owns the customer. By 2025, OTAs were driving roughly 63% of bookings at independent hotels, per the industry data reported by Asian Hospitality. We did the commission math on what that costs a hundred-room property in The OTA Tax Is Real, and the number was north of $300,000 a year at ordinary assumptions.
The pattern is worth stating cleanly, because patterns repeat before people name them. An intermediary arrives as a demand channel. It becomes a habit for the guest. Then it becomes a toll booth for the hotel. The hotel never signs a contract that says “we hereby surrender the guest relationship.” It just wakes up one decade later and finds that the guest’s default starting point belongs to someone else.
That was the first disintermediation. The question the GM asked is whether ChatGPT, Google’s assistants, and the coming wave of agentic AI are the second.
We think the phrase is worth coining, because we think it is the right frame for everything that follows: the Second Disintermediation. Same pattern, new toll booth, one crucial difference in where the toll gets collected.
The toll booth is moving from the search page to the answer.
The first disintermediation happened at the search page. The guest typed a destination, got a list of links, and the OTAs bought and optimized their way to the top of that list. The battlefield was ranked links, and hotels lost it.
What is happening now is different in kind, not just in degree. As Skift laid out in its explainer on MCP and agentic distribution, the interface is shifting from links to answers. A guest planning a trip does not browse ten results and open six tabs. She asks one question and gets one synthesized answer, and increasingly the assistant can act on that answer: check dates, compare rates, and complete a booking without the guest ever visiting a hotel website. Protocols like MCP exist precisely so that AI assistants can query live systems and transact, rather than merely summarize the web.
When the interface was links, losing meant ranking lower. When the interface is an answer, losing means not existing in the answer at all. There is no page two of a conversation.
The first disintermediation taxed how guests found you. The second one will tax how guests talk to you.
If that sentence sounds abstract, replace “talk to” with “book, modify, ask about the pool hours, request a late checkout, and plan the anniversary dinner.” The conversation is the whole relationship now. Whoever owns the conversation owns the guest.
The OTAs read this future faster than the hotels did.
Here is the part that should genuinely worry operators, because it is not speculative. It is observable, right now, in where the money and engineering are going.
Markus Busch put it more bluntly than anyone in the trade press this year: hotels are building the AI discovery layer, and OTAs are harvesting it. His argument, which matches what we see in the field, splits the AI funnel in two. The discovery layer is where the assistant learns about your property: the descriptions, the amenities, the reviews, the structured content hotels are now dutifully publishing so that AI systems can understand them. The conversion layer is where the booking actually closes: rates, availability, payment, confirmation. Hotels are pouring effort into the first layer. The OTAs are wiring themselves into the second. If that division holds, the hotel does the unpaid work of making itself legible to the machines, and the OTA collects the commission when the machine books, because the OTA’s rails are the ones the agent is connected to. The funnel’s bottom, once again, belongs to the intermediary.
The infrastructure players see it too. Aven Hospitality, the distribution platform that was SynXis in its Sabre years, announced MCP enablement across its platform with the pointed assurance that “pricing and final transactions remain under hotel control.” We read press releases the way you should read press releases: a company announcing that hotels retain control is also a company that noticed hotels are afraid of losing it. But the direction of travel is unambiguous. Every layer of the distribution stack is racing to become the pipe the agents call.
And the industry knows the stakes even where it lacks the plan. At HITEC this year, the mood had shifted from innovation to implementation, and the same refrain kept surfacing: data is simultaneously the biggest opportunity and the biggest obstacle. Which is a polite way of saying most properties cannot currently answer a machine’s questions about themselves.
An old-style data moat will not save you, because agents route around moats.
The instinctive response, especially from larger organizations, is the old playbook: hoard the data, build the proprietary integration, lock the guest into the app. Make the moat deeper.
That playbook fails here, and it is important to understand why. An AI agent is an aggregator by nature. It pulls from every source that will answer it and ignores every source that will not. Proprietary lock-in does not block the agent; the agent simply routes around you and composes its answer from whoever cooperated. Your moat does not keep the agent out. It keeps you out of the answer.
IHG understood this early, and its move is the tell. Skift reported this February that IHG is restructuring its hotel content to be machine-readable for AI agents, on the thesis that, in the coming distribution environment, not being machine-readable is the fastest way to become invisible. One of the largest hotel companies on earth looked at the agentic future and concluded that its defense was not to wall off its data but to make its data effortless for machines to consume, on its own terms, from its own source of truth.
That is the correct instinct, and it scales down. The independent hotel’s defense is the same as IHG’s: be the authoritative, machine-legible source about yourself, so the agent’s answer is built from your facts rather than from a scrape of a third-party listing last updated in 2023.
The honest caveat is that guests are not actually booking this way yet.
Now the part most vendors skip, because it undercuts the urgency slide in their deck. We will include it, because we are skeptical of hype even when the hype flatters our business.
Consumer adoption of agentic booking lags the industry conversation badly. Skift’s research this spring found that only about 2% of consumers say they want a fully autonomous AI agent booking travel on their behalf, a gap wide enough that Skift titled the piece Travel Brands Are Building AI Agents for a Consumer That Doesn’t Exist. Surveys about hypothetical future behavior are weak evidence, in both directions. But the honest read of 2026 is that discovery behavior has shifted, guests really do ask ChatGPT where to stay, while booking behavior mostly has not, yet.
We would only note that in 2003, almost nobody booked hotels online either, and the properties that treated that lag as a verdict rather than a countdown spent the next twenty years paying for it. The adoption gap is not a reason to dismiss the question. It is the grace period in which to answer it. The Second Disintermediation is not finished. It is barely started. That is precisely why this is the moment the response gets decided.
The defensible path has two parts, and a hotel controls both of them.
Everything in this series has been building to this, so let us say it plainly.
First: make your knowledge machine-readable, from a single source of truth. Rates, policies, room attributes, F&B hours, accessibility details, the answer to every question a guest or a guest’s agent could ask, structured and current and queryable. This is what survives the shift from links to answers. When the interface was a webpage, a PDF and a pretty photo gallery were enough. When the interface is an agent asking questions, the property that answers in facts wins the answer, and the property that answers in brochureware disappears from it. This is the knowledge layer we described in 99% Reliability Isn’t a Number: a live, retrievable knowledge base is not just how you make your own AI truthful, it is how you stay visible to everyone else’s.
Second: own your conversational surface. If the only place a guest can talk to your hotel is inside someone else’s assistant, you have rebuilt the OTA dependency with a better user experience and worse terms. The counter is to own the conversation where you can: your phone line, your website, your booking flow, a voice-and-visual experience that belongs to your property and closes on your rails. This is the interface argument we made in Not the Model. What It’s Wearing.: the surface the guest touches is the asset, and renting it is how the relationship gets taxed.
The two parts reinforce each other. The knowledge layer keeps you present in the conversations you do not control. The owned surface gives you a conversation you do control, one good enough that guests choose it. Feed the discovery layer, yes, you have no choice. But own a conversion layer of your own, so the harvest is not the only place the booking can land.
This is, for one paragraph and one paragraph only, what we built FlowStay to be: the machine-readable knowledge layer and the owned conversational surface, in one system, for the properties that cannot fund an IHG-sized data overhaul. One vendor’s answer. The strategy stands regardless of whose logo is on it.
The GM in that session deserved a straight answer, so here it is.
Will ChatGPT do to hotels what the OTAs did?
If a property does nothing: yes, and faster, because this time the intermediary does not need the hotel’s cooperation to describe it, only to book it, and the OTAs are already volunteering their rails for the booking. The hotel will fund the discovery layer with its content and watch the conversion layer collect the toll, and in ten years there will be a session at HITEC about winning back the agentic channel.
If a property does the two things: no. An agent-legible source of truth keeps you in the answer. An owned conversational surface keeps the close, and the guest, and the margin, yours. The Second Disintermediation is not a verdict. It is a fork, and unlike last time, the industry can see it coming.
The first toll booth was a search page. The second is a conversation.
Own the conversation.