.png)
The headlines have been breathless. AI agents are going to bypass intermediaries. OpenAI will become the new Booking.com. OTAs are finished, finita, dead. You skipped a heartbeat and started updating your resume. But hey, a new think piece declaring the travel distribution model dead lands every few months. That's just how virality works.
Get yourself a cup of tea. Let's talk about what's actually happening.
Did OpenAI spend most of 2025 building a "Buy Now" button inside ChatGPT? It absolutely did. The idea: travelers research a hotel, book it in the same chat window, never visit an OTA. Sounds like a dream for travelers. An absolute nightmare for every business that justifies its existence, employs people, and makes a living by being that intermediary layer.
In March 2026, OpenAI quietly pulled it. Travel was too complex. Users were happy to ask ChatGPT for ideas — but when it came time to enter a credit card, they left and booked somewhere they trusted. It's 2026, and we're still teaching people not to save their card details on shady websites. Not everyone listens, but cybercrime is widespread enough that people feel genuinely vulnerable about trusting sensitive financial data somewhere new.
The retreat, first reported by The Information on March 5, 2026, revealed that only around a dozen Shopify merchants had gone live with native ChatGPT checkout — out of millions — before OpenAI moved the feature into third-party apps. OpenAI confirmed the shift to Skift, framing it as prioritising discovery over direct transactions (PhocusWire). TD Cowen analysts called it a "stunning admission" that AI replacing apps as the primary commerce layer is either not happening or significantly delayed (The Silicon Review).
When the retreat was announced, Expedia's share price jumped 12%. Booking Holdings rose 8% (Tourism Tribe). The market had priced in the risk that AI would cut OTAs out. That risk, at least for now, evaporated.
So should OTAs relax? Not exactly. The real story is more interesting than either the doom narrative or the relief rally.
Agentic AI refers to AI systems that don't just answer questions — they take actions on behalf of users. In travel, this means an AI that can search flights and hotels, compare prices across platforms, and complete a booking, all without the user visiting a website.
This is different from a chatbot or a search assistant. An agentic AI acts. It transacts. It can, in theory, replace the moment when a traveler visits an OTA entirely.
The numbers reflect this shift. According to TakeUp's The Rise of AI-Planned Travel in 2026 — a survey of 300 US leisure travelers conducted in January 2026 — 90% of travelers are now aware that AI can help plan or book travel, yet only 38% have actively used it for trip planning so far (Hotel Management). A Statista report cited by Travala puts global AI usage among travelers at roughly 40% of travelers worldwide, with adoption accelerating fastest among younger travelers: 62% of Millennials and Gen Z in key markets have used generative AI for travel planning (Travala). And in the Skift Research / McKinsey joint report Remapping Travel With Agentic AI (2026), the share of travelers using AI tools "extensively" for trip planning rose 124% year over year, reaching 30% extensive usage (Skift).
The infrastructure for this shift is real and accelerating. In the past six months:
AI is genuinely good at the early stages of the travel journey. It can scan hundreds of options in seconds, compare prices across platforms, identify date flexibility windows, summarise reviews, and surface destination ideas from a half-formed brief like "somewhere warm, not touristy, good food, October." In the inspiration and research phase, AI is not just useful — it is faster and more thorough than most humans.
And then it hits the ceiling.
Complex travel planning is one of the hardest documented tasks for AI. OpenAI's most advanced model achieves roughly a 10% success rate on complex travel planning benchmarks. Earlier models managed less than 1%. Humans manage 100%.
The reason isn't raw intelligence. It's that travel is a domain built on three things AI fundamentally struggles with: ambiguity, trust, and real-world judgment.
Room mapping is a problem that sounds technical but is really about human understanding of how a global, multilingual, multicultural industry works in practice.
Supplier A calls it a "Deluxe King with City View." Supplier B calls the same physical room a "Superior Room, High Floor." Supplier C lists it as a "Premier King." And somewhere in a spreadsheet, a fourth supplier has it as "Standard Double, Non-Smoking, Upper Floors" — which in that particular property means the third floor, because the entire building has four.
An AI model can attempt to match these descriptions using language patterns. But it cannot know that "High Floor" in a boutique hotel in Prague means the fifth floor with a rooftop terrace view, while "High Floor" from a different supplier in Bangkok means floor 32 of a 45-storey tower — and that those are emphatically not the same thing. It cannot know that the "City View" in one contract was renegotiated after a taller building went up next door in NYC. It cannot know that one supplier's "Premier" is another supplier's "Standard" with a better minibar.
These are not data problems. They are judgment problems — the kind that require someone who has been in the industry long enough to know that supplier naming conventions are essentially creative fiction, and that what's written in the contract and what's in the room are sometimes only loosely related.
At Gimmonix, our verification team handles exactly this. Every booking made through RateFox is reviewed by a specialist who understands that the gap between what a contract says and what a traveler finds in the room is real — and who has the experience to close that gap before the traveler arrives.
The same applies to supplier relationships, contract nuances, and exception handling. When something goes wrong — a last-minute cancellation, a hotel closure, a policy dispute — most platforms still rely on human judgment to resolve it.
As Expedia Group's own blog put it, AI "accelerates discovery and options" but "advisors are the ones that will orchestrate the details, optimize the value, and deliver the human judgment that turns a good trip into an unforgettable one" (Expedia Group).
Luxury travel advisor Jacques Ledbetter put it more bluntly: "Clients are not paying for information. They are paying for judgment. They pay for someone who can tell them when a glossy promise won't hold up in real life" (Luxury Travel Advisor).
AI can produce fluent answers that are completely wrong. It is not the right tool for a $100,000 decision, a dietary restriction in a remote destination, or a politically sensitive itinerary.
Here is the view that gets less airtime in the doom headlines: OTAs built on clean infrastructure may end up more powerful in an agentic world, not less.
When AI agents need to shop across the travel ecosystem, they face a practical problem rarely discussed: they cannot knock on thousands of supplier doors individually. The inventory they can actually transact against is the inventory that is already aggregated, standardised, and accessible in real time.
This is precisely what Gimmonix's ecosystem is built around — a network carrying over $9 billion in buying power, connecting supply and demand through a single structured layer rather than a tangle of individual integrations. The OTAs and travel platforms that have invested in this kind of connectivity are already the layer that AI agents will route through, because they are the ones whose inventory can actually be read, compared, and transacted at machine speed.
"The ones still managing fragmented supplier connections and inconsistent data will not be bypassed by AI because of any deliberate choice — they will simply be invisible to it,"
says Gimmonix CEO Andrew Spektor.
"The RateFox model, where suppliers connect once to reach the entire Gimmonix network rather than negotiating individually with each client, points toward exactly the architecture that the agentic era rewards: structured, scalable, and clean by design."
The travel industry has been here before.
When the internet arrived, travel agents were declared finished. When OTAs arrived, traditional agents were declared finished again. When Google entered the market, OTAs were declared finished. Each time, the intermediaries that survived did so not by fighting the technology but by understanding what it could and could not do — and repositioning around the parts it could not.
The pattern holds now. Here is what that looks like in practice:
The OTAs that will lose are not the ones that failed to build a ChatGPT app. They are the ones that spent the last decade letting their supply infrastructure grow messy while focusing on marketing spend.
The ones that win will be the ones who treated clean data and deep supplier connectivity as a competitive asset — not a back-office problem.
For an AI agent to route through you rather than past you, your infrastructure needs to meet a minimum bar:
None of this is new. It is the same discipline that has always separated well-run distribution platforms from messy ones. The difference now is that the consequences of getting it wrong arrive faster — and more invisibly.
.avif)
.avif)
.avif)
.avif)
%20(1).avif)
Have questions or want to partner? Fill out the form, and our team will get back to you promptly.
.svg)
.svg){kind=small}
