How OTA Ranking Algorithms Actually Work

How OTA Ranking Algorithms Actually Work

The misconception

Ask a hotel revenue manager how Booking.com ranks properties, and you’ll usually hear some variation of: “It’s mostly price.”

It’s not. And that misconception is costing hotels both visibility on OTA platforms and, paradoxically, the strategic insight they need to reduce OTA dependency altogether.

OTA ranking algorithms are sophisticated machine learning systems that optimize for platform revenue, not for giving hotels fair exposure. Understanding how they actually work changes your entire distribution strategy. It also reveals something most hoteliers miss: the signals that determine your OTA ranking are closely related to the signals that determine whether you show up in Google, AI platforms, and every other channel where travelers discover hotels.

What Booking.com’s own research reveals

Booking.com has been unusually transparent, by OTA standards, about how its systems work. The most significant published source is “150 Successful Machine Learning Models: 6 Lessons Learned at Booking.com” by Bernardi, Mavridis, and Estevez, presented at ACM SIGKDD 2019, the top international conference on knowledge discovery and data mining.

The paper describes approximately 150 machine learning models running in production, developed by dozens of internal teams, exposed to hundreds of millions of users, and validated through randomized controlled trials. Their core finding: model performance is not the same as business performance. In other words, Booking.com doesn’t optimize for the “best” hotel for the traveler in an abstract sense — it optimizes for the outcome that generates the most value for the platform.

A companion paper, “Beyond Algorithms: Ranking at Scale at Booking.com” by Mavridis et al. (RecSys 2020), goes deeper into the ranking architecture. The system is not a single algorithm but an ensemble of specialized ML models working together. Each model predicts a different metric: click-through probability (pCTR), conversion likelihood (pCVR), and perceived quality. These models use gradient boosted decision trees in a LambdaMART framework, processing thousands of input features from user profiles, property attributes, and contextual factors.

Booking.com’s own 2024 engineering blog post confirms the architecture uses both static features — location, amenities, room types — and dynamic features recalculated in real time, including current room prices and live availability.

The key takeaway: price is one variable among hundreds. It matters, but it’s nowhere near the whole story.

The core formula: what actually drives ranking

In 2025, research from MyDataValue analysed 621 properties on Booking.com over a three-month period (January–March 2025) and developed the most detailed public model of the ranking algorithm to date. Their analysis revealed that the algorithm can be simplified to an expected profit formula:

Expected Profit = (Number of Reservations × Average Selling Price × Commission Percentage) − Acquisition Costs

This formula — also confirmed by Otamiser’s independent research — tells you everything about whose interests the algorithm serves. It is designed to maximize revenue per search for Booking.com, not to give hotels equitable exposure.

1. Conversion rate

This is the single most important factor. How often do people who view your listing actually book? A high-traffic listing with low bookings gets penalized — the algorithm interprets this as poor relevance or quality, and pushes you down.

The MyDataValue study found that conversion rate improvement contributed +1.8 points to overall ranking score — and that this compounds because higher ranking leads to more impressions, which (if conversion holds) leads to even higher ranking.

A good benchmark: properties typically see conversion rates between 2–5%. Below 2% usually signals issues with pricing, listing content, or booking friction.

2. Pricing & promotional alignment

The 2025 MyDataValue study found that over 35% of ranking factors relate to pricing and promotional settings that align with guest behaviour. This isn’t just about being cheap — it’s about pricing competitiveness relative to your market, participation in promotional programs (Genius discounts, mobile deals, last-minute offers), and rate plan flexibility.

Base ADR improvement contributed the highest individual factor at +2.8 points — but this was about optimizing price for conversion, not simply lowering it.

Crucially, dynamic pricing alone was found to address only 10–40% of the challenge in achieving optimal net RevPAR on Booking.com. The remaining 60–90% depends on non-price factors.

3. Commission tier

This is the factor hotels least like to talk about. Properties offering higher commission rates receive a measurable visibility boost. Booking.com’s own transparency page confirms this: commission level is explicitly listed as a ranking factor.

The Preferred Partner program (higher commission in exchange for priority placement) delivers an average of 65% more page views and 40% more bookings. The Genius program shows a 29% increase in bookings for participating properties. The Visibility Booster allows hotels to temporarily increase commission for higher placement.

Booking.com’s own “How We Work” page states it plainly: ranking is influenced by “how much commission they pay us on bookings” and “how quickly they usually pay it.”

4. Review score, volume & recency

In a 2025 update, Booking.com changed how review scores are calculated. Recency now matters significantly. Rather than all reviews carrying equal weight, the platform now uses a 36-month rolling average weighted toward recent reviews. Properties with strong recent scores see 15–20% higher booking conversion rates.

This means a hotel with a 9.2 score from 40 reviews in the last six months can outrank a hotel with a 9.4 score built on older reviews. Review velocity — how many new reviews you’re generating per month — has become as important as the score itself.

5. Property Page Score (content quality)

Booking.com assigns each property a “Property Page Score” reflecting the quality and completeness of its listing. This includes description detail, amenity accuracy, and critically, photos.

Booking.com’s own experiments indicate that displaying 24 or more high-resolution photos and approximately 4 photos per room type (including at least one bathroom photo) has a significant positive impact on conversion. Listings with fewer or lower-quality images consistently rank lower because their click-through rates suffer.

6. Availability & cancellation policy

Free cancellation options rank higher because they reduce booking friction — travelers are more likely to book when the perceived risk is low. Similarly, properties with availability listed further in advance (ideally 12+ months) receive more impressions, which feeds conversion data, which feeds ranking.

7. Response rate & speed

This is the factor most hoteliers miss entirely. If you’re slow to confirm bookings, frequently reject reservation requests, or have poor response times to guest messages, the algorithm flags you as unreliable. Ranking drops — not dramatically in a single instance, but quietly and cumulatively.

The personalization layer

What makes OTA algorithms particularly opaque is the personalization layer. Booking.com’s “Learning to Match” paper (Mavridis, 2018) describes how the platform builds user preference profiles with flexibility scores across multiple dimensions — dates, property type, price sensitivity, location preference.

This means the search results a traveler from Singapore sees on mobile are fundamentally different from what a traveler from Germany sees on desktop — even for the same destination and dates. The algorithm weighs search history, nationality, device type, booking behavior, and even browsing patterns within the current session.

A 2025 analysis from Hotelub confirmed that this personalization creates a ranking paradox: even if a hotel is contractually free to offer lower rates on its own website, Booking.com’s algorithm can penalize the listing by lowering its ranking if it detects rate undercutting. A Google Hotels study in 2025 found that official hotel sites are ranked lower than OTAs in 75% of cases.

Expedia: similar architecture, different emphasis

Expedia Group published its own ranking research in late 2024 via its engineering blog. Their system uses a two-stage recommendation process: first, a candidate generation step that narrows down eligible properties, then a computationally expensive ranking pass using deep neural networks.

Like Booking.com, Expedia uses learning-to-rank models trained on historical shopping data. The key differences: commission margin weighs heavily (Expedia explicitly favors listings that generate higher revenue for the platform), relevance matching is more prominent (keywords in descriptions actively help), and flexible cancellation policies rank higher across the entire Expedia Group ecosystem.

The pattern across all OTAs

Despite their differences, the major OTAs share a common optimization objective: maximize platform revenue per search. This creates a consistent pattern:

The uncomfortable implication: OTA algorithms are designed to extract maximum value from hotels, not to distribute visibility fairly. Hotels that understand this can play the game more strategically. Hotels that don’t are effectively flying blind.

What this means for your direct booking strategy

Here’s the twist that most OTA optimization articles miss: the signals that determine your OTA ranking are closely related to the signals that determine your visibility everywhere else.

• Review velocity affects your OTA ranking, your Google ranking, and whether AI platforms recommend you.
• Content quality affects your OTA conversion, your website conversion, and what AI systems extract about your property.
• Branded search volume — driven by awareness from all channels — feeds back into how OTAs personalize results for travelers who’ve already heard of you.

This is why optimizing for OTAs in isolation is a losing strategy. You’re fighting for visibility on a platform that’s engineered to extract maximum commission from you. The hotels that shift their booking mix do so not by gaming OTA algorithms harder, but by understanding that the same visibility signals — content, reviews, social presence, search authority — compound across every channel where travelers make decisions. That includes the fragmented discovery landscape we described in our analysis of SiteMinder’s 2026 traveller data.

The question isn’t how to rank higher on Booking.com alone. The question is: how visible is your hotel across the entire discovery ecosystem? Because if you’re visible everywhere, OTA dependency takes care of itself. And if you’re only visible on OTAs, you’re paying the toll forever.

---

Sources & further reading

Academic & industry papers

• Bernardi, Mavridis & Estevez. “150 Successful Machine Learning Models: 6 Lessons Learned at Booking.com.” ACM SIGKDD 2019.
• Mavridis et al. “Beyond Algorithms: Ranking at Scale at Booking.com.” ComplexRec-ImpactRS, RecSys 2020.
• Mavridis. “Learning to Match.” arXiv:1802.03102, 2018.
• MyDataValue. “Cracking the Booking.com Ranking Algorithm.” 621-property study, Q1 2025.
• Expedia Group Technology. “Choosing the Right Candidates for Lodging Ranking.” Medium, December 2024.
• Expedia Group Technology. “Channel-Smart Property Search.” Medium, July 2024.

Platform sources

• Booking.com Partner Hub. “Search results, ranking, and visibility.”
• Booking.com. “How We Work” transparency page.
• Booking.com Engineering. “The Engineering Behind High-Performance Ranking Platform.” Medium, July 2024.
• GuestTouch. “Booking.com 2025 Review Score Updates.” February 2025.

Industry analysis

• Otamiser. “How to Boost Your Ranking on Booking.com” and “Does Every OTA Determine Their Ranking Differently?” 2025.
• Hotelub. “Booking.com: A Strategic Guide for Owners.” December 2025.
• Lighthouse. “The Top 3 OTA Trends of 2025.” October 2025.