Booking.com

Booking.com's experimentation culture is a business-school case study.1 Thousands of tests a year shape every pixel of the results page. That makes it an unusually honest subject: what's on the page is there because it measurably increases bookings.

It has also drawn regulators. In 2019 the UK's Competition and Markets Authority secured commitments from Booking.com and five other sites to change how they present scarcity, discounts and paid placement.2 Persuasion on these pages is a design problem, a commercial problem and, increasingly, a legal one.

How the sample was collected

Searches were run for ten cities (five in India, five international), each for two date windows: a November weekend (20–22 Nov 2026) and a February midweek stay (9–11 Feb 2027). Every search used two adults, one room, the default “Our top picks” sort, a logged-out session, a desktop browser set to English (UK) and Indian rupees, from an Indian connection. I captured the first page of results, the top three property pages in each city and one checkout per city, and coded every message on every listing. The November searches were then repeated in four more independent browser sessions over the next ninety minutes, to test whether what a visitor sees is stable.

A single result card is the densest piece of persuasion design on the web. The example below, taken from the Jaipur sample, is typical rather than extreme. Each numbered element is a separate claim competing for the same few seconds of attention.

Pressure signals are in red; information and reassurance are in black. Notice where each sits: the red claims cluster around the price and the call to action, exactly where the decision happens.

Three things stand out. Tax display is decided by market, not by city: London and Bangkok show tax-inclusive prices, while Paris and New York don't. Scarcity varies sharply: 82% of Jaipur listings carry it, against 16% in Bengaluru and New Delhi. And the commission badge is near-universal everywhere except Bengaluru (30%).

The thumbs-up is the same yellow as the star rating and sits directly next to the hotel name. Nothing on the card says what it means. A sighted desktop user must hover over a 24-pixel icon; a mobile user must tap it, which few people do on an icon they assume they already understand.

The disclosure itself is honest. The problem is the gap between how the badge looks (an endorsement) and what it is (a commercial arrangement). By the standard of Nielsen's “match between system and the real world”,3 a thumbs-up means recommended, not pays us more.

Same search, two different first pages

Repeating the November searches in five fresh sessions exposed something the single sample hid. Every session got one of two distinct result pages. Sessions on the same version shared 71–99% of their listings; sessions on different versions shared 16%. On one version the commission badge covered 87–89% of page one, with “Preferred Plus” (the higher tier) on 27–30% of listings. On the other it covered 42%, with Preferred Plus on 4–5%.

What didn’t change is just as revealing. Tax display, scarcity and labelled ads were nearly identical across all five sessions. Struck-out prices and named deals moved with the badge, because they came with the properties. The variable was the ranking: which properties “Our top picks” chose to show first. That is consistent with a live ranking experiment. For a visitor, it means the default sort’s idea of a “top pick” depends on which version they happen to be served, and nothing on the page says so.

The contrast with explicit paid placement is telling. Listings marked “Ad” or “Featured” (6.6% of the sample) say so in visible text, and the labels are clear. Ads land in fixed slots near the top, most often position two.

Scarcity claims are the most criticised element of booking interfaces, so I tested them. On every property page sampled, I compared each “We have N left” message with the maximum quantity in that room's selector, the number of rooms Booking.com will actually let you reserve. All 54 matched.

The interesting finding is the threshold. Messages appeared for counts from one to seven and never at eight or above, even though many rooms had eight, nine or ten available. Every claim is accurate. The design choice is to announce scarcity at a level of inventory that, for most hotels, is normal.

The wording matters too: “left at this price” and “we have” describe Booking.com's allocation, not the hotel's total rooms. That's precise, and easy to misread as “the hotel is nearly full”.

The repeat sessions add one more check. Where the same hotel appeared in two sessions, it showed the same scarcity message, or the same absence of one, in 302 of 319 comparisons (95%). The rest were small moves such as “5 left” becoming “4 left”. These messages are not randomised to create urgency. They report inventory, then frame it.

A struck-out price is an anchor. It makes the real price feel like a gain. It works best when it is rare and when its reference is clear. Here it is neither: 51% of listings carry one, and only 69 of those 257 (27%) are tied to a named deal such as “Limited-time Deal” or “Late Escape Deal”.

The other 188 rely on the card's ⓘ icon to explain what “original” means. Compare EU price rules, which require any claimed reduction to be measured against the lowest price of the previous 30 days.4 A struck-out price that doesn't say what it is being compared with is a weak claim, even when it's accurate.

For this visitor, 81% of headline prices excluded tax, which was shown in smaller grey text underneath. Behavioural research calls this partitioned pricing: when a price is split into parts, people anchor on the larger part and under-weight the rest.5

The size of the gap depends on the city. In Goa or Delhi it’s about 5%. In Dubai it’s 27%, enough to change which hotel fits the budget.

The two techniques combine. A Goa resort in the sample showed ₹35,014 struck out, ₹26,190 in bold, an apparent 25% saving. The ₹5,357 in taxes underneath brings the real total to ₹31,547, so the actual saving is 10%.

One fair caveat: because tax-exclusive prices are applied consistently within a search, they reordered the price ranking in only 0.9% of listing pairs. The harm isn't in choosing between hotels on Booking.com. It is in comparing Booking.com with other sites, and with your own budget.

On a 390 × 844 phone viewport, the results page opens with the full search form you have just filled in, still expanded. Below it are sort and filter controls, and over the bottom of the screen a Google sign-in sheet. The first listing gets 24% of the screen, and its details are cut off by the sheet.

It would be easy to write this audit as a list of dark patterns. That would be inaccurate. Nothing tested here was false, and the checkout is a model of restraint that many smaller sites could learn from.

Severity is rated on how much each issue distorts the user’s decision, combining how often it appears (measured) with how misleading it is (judged). No issue is rated critical: no claim tested was untrue.

Booking.com has already done the hard part: its claims are true. What’s left is presentation. Each recommendation below keeps a working mechanism and changes only how clearly it is presented, which is where trust is won or lost.

Every message on a Booking.com card is optimised to be acted on, and only some are optimised to be understood. The scarcity count, the struck-out price and the thumbs-up all win on first read and lose on second.

Any team running conversion experiments faces the same trap. A test measures what users do in the next few seconds, not what they believe the next day. A useful rule: if a design only works when the user doesn’t fully understand it, it isn’t finished. Pair every conversion metric with a comprehension check, and the honest version of a pattern usually turns out to be the durable one.