The end of the free look: AI is killing travel’s oldest bargain

Somewhere in an airline revenue office right now, a distribution analyst is staring at a dashboard that does not make sense. Search traffic is up. Way up. Four hundred times what it was 18 months ago on some routes. And bookings are flat. Dead flat. The conversion line is a pancake. But the cloud compute invoice, the one nobody used to think about because the GDS ate that cost for 30 years, is the one number on the screen that is actually moving. It is climbing like a fever chart. And the analyst is doing the thing we all do when a spreadsheet betrays us. Refreshing it. Hoping it changes. It does not change.

What that analyst is watching is the quiet end of the oldest deal in travel. For three decades the entire industry ran on one unspoken bargain: looking was free, and booking paid for it. You could shop 100 fares, pester a travel agent, abandon three carts, and the cost of all that looking got buried inside the commission or the segment fee on the one trip you finally bought.

The look-to-book ratio in the 1990s sat around 10 to one: 10 searches, one sale. You could hide the cost of nine free looks inside one paid booking and nobody felt it. By the end of 2025, industry estimates put that ratio at 1,000 to one or worse, and the people who model this for a living think agentic search could push it to 20,000 to one, even 200,000 to one, as AI agents shop richer offers and personalise every query. That is not a trend. That is the load-bearing wall of travel economics coming out, and most of the industry is still arguing about the wallpaper.

Here is the thing the vendor decks will not tell you. The shift from human search to AI search is not a feature upgrade. It is a change in what costs money. And once you see it, you cannot unsee it, because it rewrites the math for every single player in the chain. 

The old toll booth

Think about how travel distribution actually made money before any of this. Every fee in the system was a toll collected at the moment of conversion. Not before. Global Distribution System – Amadeus and Sabre and Travelport – plumbing that connects airlines to agencies, charged the airline a segment fee when a ticket got issued. Roughly four to six dollars a segment, two and a half segments a ticket, so call it $16 to $20-some a booking. Lufthansa Group just raised those fees again for 2026, Amadeus from $19 to $21, Sabre from $24 to $26, Travelport to $26.50, per ticket.

A travel management company charged your programme a transaction fee, somewhere around $8 for an online self-service booking and $25 or $35 for one that an agent touched by phone. A hotel handed an online travel agency 15 per cent to 30 per cent of the room rate. Different numbers and platforms, same architecture: You pay when you buy.

The clever bit of it, the part nobody appreciated until it broke, is that all the looking was cross-subsidised. The GDS absorbed the computational cost of generating fares and availability. It ran the searches. It ate the compute. The airline only paid on the booking. So, the airline did not care if you searched once or searched 50 times, because the search was somebody else's problem and the booking was where the meter ran.

The look-to-book ratio could climb to 300 to one and the model still held, because one fat commission could still bury 300 cheap looks. The whole system was an all-you-can-eat buffet where roughly one diner in 300 quietly paid the check for the entire dining room. It worked because human beings get tired. You shop, you get bored, you book, you go to dinner. Human attention was the rate limiter that made the free-search economy solvent.

AI agents do not get tired. They do not get bored. They do not go to dinner.

The new meter 

Now flip it. An agentic AI does not search the way you do. It does not glance at the first page of results and give up. It reasons, it breaks the task into steps, it calls a model 10 or 20 times to book a single trip, it checks its own work, it second-guesses, it re-queries. Gartner's March 2026 analysis found agentic workflows burn somewhere between five and 30 times the tokens of a plain chatbot answer for the same task. Tokens are the new segment fee, except they are charged on the look, not the book. 

Yes, tokens got cheaper. Input pricing collapsed from around $30 per million tokens in mid-2023 to under $3 for comparable capability by early 2026. That sounds like the problem solves itself. It does not. Because consumption is outrunning the price drop, and it is not close. Gartner expects per-token inference costs to fall by more than 90 per cent by 2030 and total inference spend to rise anyway, because volume is exploding faster than unit cost is falling. An unconstrained agent can run $5 to $8 in compute to grind through a single complex task. Multiply that across every business traveller whose AI assistant fires 4,000 queries to plan one offsite, and the buffet model does not just strain. It detonates. 

The cost of travel shopping used to live on the booking, where it was rare and recoverable. AI moves the cost onto the search, where it is constant and recoverable from nobody. The old fee was a toll on conversion. The new cost is a tax on curiosity. When curiosity is a machine, it is infinite. 

So, the cost did not vanish. It moved. It came off the booking, where it was rare and recoverable, and settled onto the search, where it is constant and, for now, recoverable from no one. That phrase, "for now," is the whole story. A cost that lands on nobody never stays on nobody. It rolls downhill until it finds a desk to sit on, and in this chain there are five desks: the airline, the GDS, the TMC, the traveller, and the one you are sitting at.

The reckoning has already started, and it is not landing evenly. And there is a serious counterargument coming, that none of this is inevitable, that the explosion is a design choice someone is making on your behalf rather than a law of physics.  

In part two, we walk the chain and find out who actually eats it. One of the answers is you, the travel buyer.

______________________________________________________________________

Check out part two on Tuesday, 23 June. 

Steve Clagg is the founder of Clagghaus Consulting, where he works with corporate travel buyers and procurement leaders on technology strategy, supplier evaluation, and programme architecture. If your organisation is navigating enterprise AI integration and trying to figure out where your travel programme fits, that is exactly the kind of work Clagghaus does.