The traveller Booking.com was built for
Booking.com's search form has four fields. Together they encode assumptions about who travels, why, where they begin, what they want to know, and which currency they think in. Most of the world's travellers fail the schema before they fail the search.
Key stat
people, on a conservative count, whose home currency does not appear in Booking.com's dropdown. The platform offers 52 currencies. The world uses roughly 180. Nigerian naira, Bangladeshi taka, Vietnamese dong, Kenyan shilling, Ghanaian cedi: not on the list. They convert mentally on every search.
You open Booking.com. The search bar wants four things from you: a destination, a check-in date, a check-out date, and a count of rooms and guests. There is a small button below for "I'm flexible." There is no other field. Once you fill those four, the site has everything it thinks it needs to plan your trip.
Now picture the trip in your head. You want to spend October flying from Johannesburg to Bangkok, then on to Hanoi, then Jakarta. You are visiting friends in two of those cities and going to a wedding in the third. You haven't fixed dates because the wedding has not been confirmed. You travel on a South African passport. It opens 100 countries without a visa[1], which puts you in the stronger half of African passports (Seychelles and Mauritius both rank higher), but at roughly half the access of a Singaporean passport, which opens 192. You need to know which of the three Asian cities you can fly into without paperwork this month. You'd prefer guesthouses owned by people from the local community where they exist. You want food that meets your household's restrictions within walking distance. Your budget is in rand, and the dollar conversion shifts every day.
You start filling the form. Destination: you pick one of three cities, because the form does not understand a circuit. Dates: you choose a range and lie about the wedding. Rooms: one. Guests: one, though you will be eating with seven.
You click search.
What comes back is a list of Bangkok hotels sorted by something the platform calls "Our top picks." The currency at the top right is set to whatever the platform's geolocation thinks your money is. Visa is not mentioned anywhere on the page. None of the things you actually care about is a filter: not community ownership, not dietary need, not your friend's wedding, not your aunt's prayer schedule.
You have not searched. You have submitted to a schema.
The claim
The online travel agency, or OTA, is not a neutral marketplace. (Booking.com, Expedia, Agoda, Hotels.com, Trivago, and Airbnb are all OTAs, or close enough that the schema is the same.) It is a Western leisure-travel product wearing a global brand. Its schema, the four fields you just filled, encodes assumptions about who travels, why they travel, where they begin, what they want to know, and which currency they think in. The defaults are not technical conveniences. They are decisions, made when Bookings.nl launched from a student's bedroom in Enschede in 1996, and ratified again every time another OTA copied the form. Most of the world's travellers fail the schema before they fail the search.
This essay is about how that schema works, who it works for, and what gets lost in the conversion. The receipts that follow come from a live audit of Booking.com on 27 May 2026.
The audit
Five things the search bar does that nobody writes down. The form is small. Everything it does follows from how small it is.
1. The four fields
Booking.com's homepage search bar accepts exactly four inputs. A destination, entered as a single text field. A check-in date and check-out date, entered as a single range. An occupancy counter that defaults to "2 adults · 0 children · 1 room." And a search button.
The default is a couple. Not a solo traveller, not a parent with a child, not three generations, not seven people eating from one host's kitchen. The category itself, "occupancy," is the platform's frame for who you bring with you, and the default state inside that category is two adults sharing a single room. Expedia's homepage uses the same four fields. Airbnb's is slightly more flexible: it lets you skip the dates. The destination remains a single field. The schema is industry-wide.
The form does not understand a circuit. It cannot accept "Bangkok, then Hanoi, then Jakarta" as a single search. You must run three searches, or you must abandon two of the cities. It does not understand an open date. It cannot accept "October, not sure exactly which week." You must pick a range and adjust later. It does not understand a sleeping arrangement that is not "rooms in this property." If you are staying with family but eating in restaurants, the form has no field for the part of the trip that costs money but does not involve a room.
2. The currency dropdown
Booking.com offers 52 currencies. The world uses roughly 180.
Eight currencies are featured at the top of the dropdown. When I ran the audit, the order, top to bottom, was EUR, USD, GBP, AED, SGD, AUD, JPY, PHP. The eighth slot is geo-personalised; the platform had resolved the test IP to the Philippines and bumped PHP up. The remaining 44 currencies appear alphabetically below.
Selected positions in the alphabetical list: BRL at 12, EGP at 19, INR at 25, IDR at 26, MXN at 33, ZAR at 45, KRW at 46, THB at 49.
A partial list of populous economies whose currency is not in the dropdown at all: the Nigerian naira (220 million people), the Bangladeshi taka (170 million people), the Vietnamese dong (100 million people), the Kenyan shilling (55 million people), the Ghanaian cedi (33 million people). On a conservative count from just those five, around 580 million people have a home currency that does not appear on the platform. Add the unverified absences across the rest of Africa and Southeast Asia and the number passes 600 million. They convert mentally on every search. The platform sells them rooms in dollars and they pay their bank to settle in the currency the platform never asked them about.
- EURposition 1
- USDposition 2
- GBPposition 3
- AEDposition 4
- SGDposition 5
- AUDposition 6
- JPYposition 7
- PHPposition 8
- BRLposition 12
- EGPposition 19
- INRposition 25
- IDRposition 26
- MXNposition 33
- ZARposition 45
- KRWposition 46
- THBposition 49
- NGNNigerian naira
- BDTBangladeshi taka
- VNDVietnamese dong
- KESKenyan shilling
- GHSGhanaian cedi
52 currencies offered out of roughly 180 in circulation. Position 8 geo-bumped to PHP because the audit was run from a Philippines IP.
3. The filter wall
A Bangkok search returns 22 filter groups and 115 filter items. Counting the ones the platform offers is the easy part. The reveal is in what it does not.
What you can filter for: price per night (in the currency you selected), property type (hotel, apartment, ryokan, hostel, resort, B&B, guest house, capsule hotel, villa, motel, holiday home, homestay, lodge, chalet, holiday park, boat), bedrooms, bathrooms, review score, free cancellation, no prepayment, breakfast included, all-inclusive, breakfast and dinner included, self catering, swimming pool, parking, free wifi, airport shuttle, fitness centre, private bathroom, air conditioning, balcony, washing machine, kitchen, star rating from 1 to 5, neighbourhood, distance from the city centre, hotel chain brand, double or twin beds, sauna, massage, happy hour, pets allowed, adults only.
What you cannot filter for: visa requirement, halal food, halal-certified accommodation, kosher, jain, vegetarian or vegan dietary options, prayer room, qibla direction, community-owned guesthouse, locally-owned versus international chain, women-only floors, women-only pool hours, women-safe at night, minority-owned or indigenous-owned, walking distance to a mosque or temple or church, public transit access for travellers who do not drive, local-language reception staff, currency-stable pricing, family group larger than six, three-generation booking, modest-swimwear sections.
The audit page contains no mention of the word "visa," "halal," "kosher," or "prayer." Anywhere on the page.
The filter group called "Travel group" exists. Its two options are "Pets allowed" and "Adults only." There is no "with kids" or "with parents" or "extended family." The category itself reveals what the platform thinks a travel group might be.
- Price per night
- Property type (hotel, hostel, ryokan, capsule, villa, motel, boat, more)
- Bedrooms and bathrooms
- Review score (Booking's composite)
- Free cancellation, no prepayment
- Breakfast, all-inclusive, self catering
- Pool, parking, free wifi, airport shuttle, fitness centre
- Private bathroom, air conditioning, balcony, washing machine, kitchen
- Star rating (1 to 5)
- Neighbourhood and distance from centre
- Hotel chain brand
- Bed preference (double, twin)
- Sauna, massage, happy hour
- Pets allowed, adults only
- Accessibility (grab rails, roll-in shower, hearing accessible)
- Visa requirement
- Halal food or hotel certification
- Kosher, jain, vegetarian dietary needs
- Prayer room, qibla direction
- Community-owned or locally-owned
- Women-only floors or pool hours
- Women-safe at night
- Minority-owned or indigenous-owned
- Walking distance to a mosque, temple, synagogue, church
- Public transit access for travellers who do not drive
- Local-language reception
- Currency-stable pricing
- Family group larger than six
- Three-generation booking
- Modest swimwear or family swim hours
115 filter items across 22 groups. The words visa, halal, kosher, and prayer appear zero times anywhere on the audited page.
4. "Our top picks" is a paid position
The default sort on every Booking.com search results page is called "Our top picks." It is not a recommendation.
Booking.com charges hotels a commission of roughly 10 to 25 percent of the room rate, depending on country, property tier, and visibility programs. Genius and Preferred Partner programs add roughly 3 to 5 percent on top in exchange for placement[2]. The platform does not publicly publish the algorithm, but its own partner-facing material confirms that visibility-program tier, commission rate, and conversion data are inputs to the default sort.
The hotel at the top of "Our top picks" is the one that pays the platform the most and that the platform has the most data on. The label is a recommendation in language and a paid position in mechanism.
- 10–25%Booking.com base commission
- 3–5%Preferred Partner or Genius placement premium
- 1–3%Payment processing
- 67%Hotel revenue
Visualised at the upper bound of each band. Lower-bound scenario leaves the hotel around 86 percent of the room rate.
5. Smart filters does not change the schema
Booking.com has recently added a "Smart filters" panel above the filter wall. Its prompt is "What are you looking for?" It accepts natural language.
The catch is that it does not change the underlying schema. Smart filters helps you navigate the 22 groups and 115 items that already exist. It surfaces filters you might not have found. It does not surface filters that do not exist. Ask it for a halal-friendly hotel near a mosque and it returns the closest approximation it can express, which is nothing.
The AI layer on top of Booking is the most visible part of where this is going. The schema underneath is the part that decides whether the AI is useful.
The machinery
How the schema got this way is a sixty-year story.
Bookings.nl was founded in 1996 by Geert-Jan Bruinsma, then a University of Twente student, with the server running under his desk in Enschede. The four-field schema reflected the European leisure traveller it served: a person on a fixed annual holiday flying from one country to another, sleeping in a hotel, and eating where the hotel suggested. That was the market. The product fit it. Merged with Bookings Online in 2000, the company rebranded as Booking.com and moved its headquarters to Amsterdam.
Below Booking sits the Global Distribution System layer. Sabre, developed by American Airlines and IBM, was initially deployed in 1960 and fully operational across AA's network by 1964[3]. Amadeus was founded on 21 October 1987 by Air France, Iberia, Lufthansa, and SAS, each holding 25 percent, as a European counterweight to American-dominated GDSs. Travelport, descended from Galileo and Worldspan, traces the same era. These systems were built for airlines and travel agents decades before consumer internet. They set the data schema for airline and hotel inventory. An OTA can only offer a filter the underlying data contains. Visa status is not in the GDS. Halal-friendliness is not in the GDS. Community ownership is not in the GDS.
Priceline acquired Booking.com for $133 million in July 2005[4]. The schema did not get redesigned. It got scaled. Booking Holdings today owns Booking.com, Priceline, Agoda, Kayak, OpenTable, and Rentalcars.com, all riding the same four-field shape. Expedia Group rides parallel infrastructure on the American side: Expedia, Hotels.com, Vrbo, Travelocity, Orbitz, Hotwire, and a 61.6 percent majority stake in Trivago. Two holding companies, the same product, billions of users, one schema.
Commission economics keeps the schema in place. The platform takes 10 to 25 percent per booking. To make that math work, the platform needs volume, and volume requires standardisation. Adding a "community-owned" filter requires hotels to enter that data. A two-room guesthouse in Lagos or Jakarta will not enter custom structured data without payment, and there is no payment because the platform's margin model does not allow it. The schema is held in place by economics. The team in Amsterdam can see what is missing. The model does not let them fix it.
What would have to change is one of two things. Either hotels publish more structured data, which requires either a state mandate or a platform with enough leverage to demand it. Or the platform infers the missing dimensions from unstructured signals like reviews, photos, web mentions, and third-party datasets. Booking can build the second option. They have not.
The pattern
The schema failure is not universal. When the default does not fit a market, the market builds its own platform.
Trip.com Group. Founded as Ctrip in Shanghai in 1999. The company reported revenue of RMB 53.3 billion (approximately $7.3 billion) for the full year 2024, up 20 percent year-on-year[5]. Trip.com Group owns Ctrip, Trip.com, Skyscanner, and Qunar, and holds a minority stake in MakeMyTrip. The platform serves an audience that Booking.com does not.
IRCTC. Built by the Indian government in 2002, the Indian Railway Catering and Tourism Corporation runs the rail booking system for a country with roughly 7,000 stations. In financial year 2024-2025, IRCTC averaged 1.39 million tickets per day, with digital channels handling 86.38 percent of all reserved tickets[6]. The product is built around how India travels by rail. No global OTA was set up to express it. The country built its own.
HalalBooking and Nusuk. HalalBooking was registered in the UK in 2010 (the company markets a 2009 founding, and its current entity was relaunched in 2014), designed for travellers who need alcohol-free hotels, women-only pool hours, qibla direction in the room, and halal restaurants nearby. Nusuk, the Saudi government's official Hajj and Umrah booking platform, launched on 26 September 2022, succeeding the Eatmarna app. Nusuk handles the Mecca quota system, group registrations from a single mosque, and the specific schema of a religious circuit. The Saudi General Authority for Statistics recorded 1,833,164 Hajj pilgrims in 2024, of which 1,611,310 were international[7]. None of them booked through Booking.com.
The shape that repeats: when the default schema does not fit the market, the market builds its own. Each of these platforms encodes assumptions about who travels and what they need to ask. The assumptions are different from Booking's. They fit their users in ways the default OTA does not. The point is not that any one of them is the answer. The point is that the schema is a choice, and other people have made other choices.
- Destination (single)
- Check-in date
- Check-out date
- Occupancy (rooms + guests)
Audited 27 May 2026.
- From station
- To station
- Journey date
- Class of travel
- Quota (general, Tatkal, ladies, more)
From the live booking interface at irctc.co.in.
- Pilgrim nationality
- Pilgrimage type (Hajj or Umrah)
- Package level
- Travel dates
- Group size
- Hotel proximity to the Haram
From nusuk.sa booking flow.
The platforms describe different trips. Each schema is a choice about who the platform was built for.
The implication
On 6 October 2025, at OpenAI's DevDay, OpenAI announced its Apps SDK, which lets third-party services run inside ChatGPT conversations. Booking.com was a launch partner, alongside Expedia, Canva, Coursera, Figma, Spotify, and Zillow[8]. The same four fields, now reachable through a chat box.
The agent layer is the next top of the funnel. The question is what schema it inherits.
The training data is the previous layer's product. The web that large language models read for travel is the Booking, Expedia, Airbnb, TripAdvisor, and Google Travel corpus, in English, with the defaults that those platforms set. When the agent layer ships with the previous layer's data, it ships with the previous layer's defaults. The schema reproduces itself by way of the dataset.
The break, if it comes, is also visible. Agents can infer structured data from unstructured signals in a way OTAs cannot. They can read a forum thread, a travel writer's caption, a comment about a women-friendly hostel, and synthesise an answer the filter set could not produce. Whether the next-generation travel layer breaks the schema or repeats it depends on who builds it, what they train it on, and which traveller they design for.
The traveller Booking.com was built for has been served for thirty years. The next default is being designed now. The four fields are not neutral. They never were.
A note on method
The audit findings in this essay come from a live walk-through of Booking.com on 27 May 2026, covering the homepage form, the currency selector, and a Bangkok search results page. Filter counts, currency positions, and the wording of category labels are taken directly from the platform as it rendered that day. Where the piece describes a platform's behaviour without an inline citation, the description is drawn from the platform's own product surface. Statistical and historical claims are from the named primary sources below. Platforms iterate; specific positions and counts may have shifted by the time you read this. The structural argument does not depend on any single number.
Sources
- [1]Henley Passport Index, February 2026 update. South Africa ranks 47th with 100 visa-free destinations. Singapore ranks 1st with 192. Japan and South Korea tie at 2nd with 188.↑
- [2]Booking.com does not publish a single global commission rate card. Industry analyses of partner-facing materials place the base rate at roughly 10 to 25 percent, with Preferred Partner and Genius placement adding around 3 to 5 percent on top. See Lodgify's guide to Booking.com fees and Preno's commission analysis.↑
- [3]IBM, Sabre system history. Formal American Airlines + IBM agreement signed 1957. Initial experimental installation 1960 at Briarcliff Manor, NY. Fully operational across American Airlines's network by 1964.↑
- [4]Skift, Oral history of the Booking.com acquisition. Priceline acquired Bookings B.V. (the parent of Booking.com) for $133 million in July 2005.↑
- [5]Trip.com Group Limited, Q4 and Full Year 2024 Financial Results, released 24 February 2025. Revenue of RMB 53.3 billion (~$7.3 billion), up 20 percent year-on-year.↑
- [6]IRCTC Annual Report 2023-24; Indian Railways, Annual Report and Accounts 2024-25. IRCTC averaged 1.39 million tickets per day in FY2024-25, with digital channels handling 86.38 percent of all reserved tickets.↑
- [7]Saudi General Authority for Statistics, Hajj Statistics 2024 publication. 1,833,164 Hajj pilgrims in 2024, of which 1,611,310 were international.↑
- [8]OpenAI, Introducing apps in ChatGPT, announced 6 October 2025 at DevDay. Booking.com, Expedia, Canva, Coursera, Figma, Spotify, and Zillow named as launch partners. See also Skift's coverage of the announcement.↑
Get the next essay in your inbox.
Long-form on travel, AI, and the people the platforms were not built for. One email a week. No noise.