Hotelist: Why did Pieter Levels build an Airbnb alternative?

Is Hotelist actually an Airbnb alternative?

Hotelist is currently much closer to an independent hotel search engine than a real Airbnb alternative.

Hotelist does not handle hosts, reservations, payments or a two-sided accommodation marketplace. The product sits earlier in the trip: it tries to answer the annoying question of which hotel is actually worth booking.

That difference is easy to see from the product today. Hotelist currently covers roughly 90,500 hotels across more than 10,500 cities and 183 countries. Travelers can search by price, hotel age, chain and unusually specific features such as a real weightlifting gym, a usable work desk or a kitchen. Each property receives a Hotelist Score built from several sources instead of simply displaying the rating from Booking.com or Google.

The Airbnb comparison still makes sense because Levels has become increasingly critical of the large platforms people use to choose accommodation. But Hotelist is attacking their recommendation layer. Booking can still happen elsewhere.

Product	What it mainly does	Where Hotelist overlaps
Airbnb	Finds and books accommodation	Reviews and accommodation discovery
Booking.com	Finds and books hotels	Hotel rankings, ratings and search
Google Maps	Helps people discover places	Reviews and local discovery
Hotelist	Tries to identify the best hotels	Independent ranking and hotel research

Where did Pieter Levels get the idea for Hotelist?

The idea for Hotelist came unusually directly from Pieter Levels and Lex Fridman complaining about bad hotel search during their podcast conversation.

Fridman described a problem many frequent travelers recognize: hotel listings often tell us almost nothing about the details that actually affect a stay. Does the room have a proper desk? Is it quiet enough to work? What does the room really look like? Is the advertised gym useful?

Levels immediately related to it. He said he had learned that even hotel chains were unreliable shortcuts because one property could be excellent while another hotel from the same chain was bad. Room quality could also vary inside the same property.

Fridman then suggested having the kind of detailed data Nomad List provides for cities, except for hotels. Levels said he had already thought about the idea.

Hotelist now explicitly credits that podcast conversation as the interview that inspired the product.

So we have something better than a retrospective founder story here. We can actually see the problem being discussed before Hotelist existed.

Did Pieter Levels build Hotelist because of one bad Airbnb experience?

The recent Airbnb dispute gave Pieter Levels a reason to push Hotelist much harder, but Hotelist had already been conceived long before that happened.

Levels recently said Airbnb removed a negative review that he and his partner had written about a host they described as threatening, while the host's negative review of them remained online. Around the same time, Levels wrote that he had built Hotelist because Airbnb, Google Maps and booking websites remove negative reviews until almost everything ends up around 4.7 stars.

That is Levels' version of what happened, and we should separate it from Airbnb's own rules. Airbnb currently says hosts and guests can dispute reviews, but a negative opinion or disagreement over a star rating is not enough to have one removed. Reviews can disappear when they violate rules covering issues such as retaliation, irrelevant content, fake experiences or prohibited material.

That distinction matters: the broad claim that Airbnb simply deletes bad reviews goes further than the evidence we have.

What we can establish much more confidently is timing. Levels and Fridman were already discussing essentially the same hotel product years before this dispute. The Airbnb episode revived the argument behind Hotelist rather than creating it.

What does Pieter Levels think is broken about hotel search?

Pieter Levels thinks today's hotel search combines three bad problems: inflated ratings, commercially influenced rankings and surprisingly weak information about what a hotel is actually like.

Hotel ratings are the most visible problem. If almost every acceptable property is somewhere around 4.4, 4.6 or 4.7 out of 5, those numbers look precise while telling us relatively little. A traveler still has to work out whether a 4.6 is genuinely excellent or merely normal for that platform.

Then comes ranking. Booking websites have good reasons to show properties that are likely to convert into bookings, and commercial relationships can enter that system too.

Finally, structured hotel data remains crude. A property either has a gym or does not. Yet a traveler who lifts weights cares enormously about whether "gym" means a squat rack and dumbbells or one treadmill in a small room. The same problem appears with desks, kitchens, room condition and even the age of the interior.

Hotelist tries to solve those three problems together. That combination explains the product better than Levels simply being unhappy with Airbnb.

Is Booking.com really pay-to-play?

Yes, commercial payments currently affect visibility on Booking.com, although calling the whole Booking.com ranking pay-to-play would go too far.

Booking.com's own explanation of how the platform makes money is unusually clear. Accommodation providers pay commission when a stay is booked. Properties in the Preferred Partner Programme pay a higher commission. Booking.com also says properties displaying an "Ad" badge have paid to appear there.

European regulators have looked more closely at the Preferred system. Italy's competition authority investigated Booking.com's practices after raising concerns that Preferred and Preferred Plus properties could receive greater visibility in search results while paying higher commissions.

An earlier Italian competition case described Preferred Partner and Preferred Plus as programmes offering search visibility benefits in exchange for higher commission payments and competitive pricing commitments. The investigation was later closed after Booking.com offered commitments.

So Levels has a solid factual basis when he says money can affect hotel visibility. Where his rhetoric becomes too broad is when every Booking.com result is treated as if the highest bidder simply buys first place. Booking.com uses several ranking factors, and quality, availability, user behavior and conversion also influence results.

Booking.com feature	Commercial relationship	Can visibility change?
Normal listing	Commission after a booking	Ranking depends on several factors
Preferred Partner	Higher commission	Yes, greater visibility can be part of the programme
Preferred Plus	Higher-tier commercial programme	Yes, additional visibility can be offered
Sponsored listing	Direct payment for advertising	Yes
Hotelist	Currently says hotels cannot pay for ranking	No paid placement claimed

Are online hotel reviews really broken?

Online hotel reviews have real weaknesses, but the evidence supports a narrower criticism than saying most reviews are fake or routinely censored.

Fake-review manipulation is serious enough that the U.S. Federal Trade Commission introduced a specific rule covering purchased fake reviews, sentiment-dependent incentives and certain forms of negative-review suppression. The FTC did not create that rule specifically for hotels, but it confirms that review manipulation is a large enough online-market problem to require dedicated enforcement.

Platforms are also fighting the problem themselves. Airbnb currently runs systems that try to identify reviews that may not come from genuine stays and removes reviews that violate its policies.

For Hotelist, rating compression may be the bigger problem. We do not need most reviews to be fraudulent for a rating system to become less useful. If almost every decent hotel accumulates a very high score, finding the genuinely exceptional properties becomes difficult.

Hotelist's answer is therefore quite aggressive: keep the underlying reviews, then reinterpret what their scores really mean.

How does Hotelist score hotels differently?

Hotelist currently builds its hotel rankings from conventional ratings, wider traveler commentary and AI analysis of hotel photos.

First, Hotelist gathers scores from several booking and review platforms. Those raw ratings are then normalized because different platforms use their rating scales differently.

Hotelist also searches material outside conventional review pages, including forums, travel communities and trip reports. AI extracts recurring positive and negative experiences from that material.

Then vision models examine hotel photos. Hotelist says these models evaluate things such as cleanliness, how new a property looks, amenities and general atmosphere.

Finally, Hotelist displays a consensus measure showing how closely the different sources agree. A property with a high score but low consensus deserves more investigation than one that is rated strongly across several independent inputs.

The result is a much more opinionated rating system than a normal review average.

Hotelist input	What Hotelist tries to learn
Booking and review ratings	How travelers scored the hotel
Normalized scores	Whether that rating is genuinely unusual on that platform
Wider web discussion	What travelers repeatedly praise or complain about
Hotel photos	Condition, style and visible amenities
Consensus	Whether different sources broadly agree

Why does Hotelist change the ratings it collects?

Hotelist normalizes hotel ratings because a 4.6 out of 5 tells us very little without knowing what scores other hotels receive on the same platform.

Imagine a review site where almost every hotel sits between 4.2 and 4.8. A hotel rated 4.5 looks excellent on a five-point scale, yet it is roughly in the middle of the real distribution.

Hotelist stretches the range that each source actually uses onto its own 0-to-10 scale. Its current explanation says a typical hotel on a heavily inflated platform can therefore land near 5, while a genuinely rare top score moves closer to 10.

We can see the effect in Hotelist's live dataset. Its global average is currently around 7 rather than clustering close to the theoretical maximum. At chain level the spread becomes much wider: Mandarin Oriental and Four Seasons currently sit around the mid-8s, while Motel 6 and Extended Stay are below 4.

Those numbers create separation that conventional five-star averages often hide.

Normalization will not tell us whether Hotelist is correct about a particular hotel. It does make the ranking much more discriminating.

What can Hotelist's AI tell us that normal hotel filters cannot?

Hotelist's most useful AI feature may be its ability to answer questions that hotel databases usually reduce to a useless yes-or-no field.

Take gyms. Booking sites can tell us whether a property claims to have fitness facilities. Hotelist tries to inspect the images and determine whether there is actual weightlifting equipment.

The same approach works for work desks, kitchens, bathtubs, room condition and interior style. Those are small details until one of them determines whether a week-long stay works.

Photos are particularly useful here because hotel descriptions are structured around what the property says it offers. Images provide another source of evidence.

As seen above, Hotelist also combines those visual judgments with conventional ratings and traveler commentary. The interesting part is the combination: a hotel can look great in professional photos while repeated traveler comments complain about noise, or receive strong reviews while its supposedly serious gym turns out to be tiny.

That gives Hotelist a level of detail that normal booking filters still struggle to provide.

Why is Hotelist suddenly possible for one developer?

Today's AI models make Hotelist much cheaper to build because software can now read and classify the messy information that previously required human researchers.

Hotelist currently spans roughly 90,500 properties. A human editorial team manually inspecting photos, reading discussions, checking amenities and writing summaries for that many hotels would be expensive.

Language models can process large amounts of traveler commentary. Vision models can classify room photos. Software can then combine those outputs with structured rating and pricing data.

There will be errors. An old hotel photo can be mistaken for the current room. A recently renovated property may have stale information online. An AI model can confuse design style with cleanliness or miss equipment hidden outside the images it receives.

But the economics have changed dramatically. Levels can attempt something that used to look more like a media company or travel-guide operation while remaining a very small project.

That timing helps explain why an idea discussed years earlier has become a much more ambitious product now.

Is Hotelist basically Nomad List for hotels?

Yes, Hotelist increasingly looks like Pieter Levels applying the Nomad List formula to individual hotels instead of cities.

Nomad List grew from Levels wanting better information about places where he could live and work. It turned messy questions about cities into sortable attributes: cost, internet, weather, safety and dozens of other variables.

Hotelist follows the same product instinct. Travelers can move beyond "Is this hotel highly rated?" and ask for properties that are new, affordable, suitable for work or equipped with particular amenities.

That also explains why Hotelist feels more like a dense database than a polished online travel agency. Levels has spent years building products where the advantage comes from taking scattered information, structuring it and making it easy to filter.

Hotels are an obvious extension of that idea because accommodation has been part of his life for years. During the Lex Fridman conversation, Levels was already talking from experience about inconsistent hotel chains and bad room information.

Hotelist fits his old playbook surprisingly closely.

Why is Pieter Levels spending money on Hotelist if it is free?

Levels is currently treating Hotelist as a product he personally wants to exist, even though he says the site costs him thousands of dollars per month.

He says Hotelist does not currently make money and does not use affiliate commissions. The running costs come from gathering and processing information through APIs and web scrapers.

That choice is more believable in Levels' case because he already has profitable businesses and has repeatedly argued that every project does not need its own business model.

There is also a useful recent test of whether he genuinely uses the thing. Levels said he found Woo Suites in Athens through Hotelist and then stayed there, describing the property positively afterward.

For now, Hotelist has passed the founder's own test: it helped him choose a hotel he liked.

That is a small sample, obviously. But it is more informative than launching a travel-ranking product that the founder never actually uses.

Can Hotelist make money without ruining the idea?

Hotelist could make money fairly easily in travel, but the obvious business model would create exactly the conflict Pieter Levels is criticizing.

Affiliate booking commissions are the obvious route. A traveler discovers a hotel on Hotelist, clicks through to a booking service and Hotelist receives a fee if that person books.

Levels and Fridman even discussed affiliate booking revenue when they first talked about the hotel idea.

The problem is incentive design. Once one booking partner becomes more profitable than another, Hotelist has to prove that its rankings remain independent from the economics of the transaction.

Other options fit the product better. Hotelist already exposes an MCP interface that allows AI systems to query its hotel information. A proprietary hotel dataset could eventually be sold through API access, premium search tools or licensing without changing which hotel ranks first.

There is no urgent need to solve this today if Levels genuinely accepts the running cost. But monetization is where Hotelist's independence claim would face its hardest test.

Does Hotelist have anything competitors cannot easily copy?

Hotelist's AI features are easy to copy individually; the harder asset would be a trusted hotel dataset built from years of independent analysis.

Booking.com, Expedia, Google and Airbnb all have access to strong AI models. None needs Pieter Levels' code to summarize reviews or analyze images.

Hotelist therefore has very little technological protection from AI alone.

A more durable advantage could come from continuously building structured information that the large platforms do not expose cleanly: normalized cross-platform scores, verified amenities, photo-derived attributes, recurring complaints and source consensus for tens of thousands of hotels.

Trust could become the other advantage. If travelers gradually believe Hotelist rankings more than commercially influenced search results, copying a vision model would not automatically reproduce that reputation.

Hotelist has not earned that kind of trust yet. But if a moat develops, data and trust are much more plausible candidates than AI itself.

What's the biggest problem with Hotelist right now?

Hotelist still has to prove that its own algorithmic opinion of a hotel is more accurate than the imperfect ratings it criticizes.

That is the uncomfortable part of the product.

Hotelist takes ratings from other sources, transforms them, mixes in web commentary and adds judgments from AI models. Every extra source can improve the answer, but every transformation can also introduce another mistake.

The photo analysis is particularly subjective. A vision model may interpret an older interior as worse quality even when the room is spotless. Online discussions may overrepresent travelers who had unusually good or unusually bad experiences. Normalization can amplify small rating differences.

Hotelist's consensus score helps because it makes disagreement visible. The site has also started showing a small set of hotels that were personally reviewed, which gives us another comparison point. On the current Stats page, those human scores sometimes line up closely with Hotelist and sometimes diverge noticeably.

That is useful honesty, but it also shows why Hotelist still needs validation.

The real test is simple: over hundreds or thousands of stays, do people choosing highly rated Hotelist properties end up happier than people following normal Booking.com, Google or Airbnb rankings? We do not have that evidence yet.

Could Hotelist seriously compete with Booking.com or Airbnb?

Hotelist can realistically compete for the hotel-discovery decision, while competing with Booking.com or Airbnb for the entire booking transaction would require a completely different company.

Booking.com brings live inventory, prices, reservations, cancellations, payments, hotel contracts, support and enormous existing demand. Airbnb has its own supply network, payments infrastructure, host relationships and proprietary stay data.

Hotelist currently avoids nearly all of that complexity.

That narrower scope may actually be an advantage. Levels can focus on answering one question better: "Where should I stay?"

A traveler could eventually use Hotelist to decide that a particular hotel is best and still open Booking.com to reserve it. Hotelist wins influence over the decision without having to rebuild the global travel infrastructure underneath it.

That is a much more believable path than Hotelist becoming another full online travel agency.

Layer	Hotelist today	Booking.com / Airbnb
Hotel discovery	Yes	Yes
Independent scoring	Core product	One part of a larger platform
Live inventory	Limited / external	Core infrastructure
Reservation processing	No	Yes
Payments and cancellations	No	Yes
Global supplier marketplace	No	Yes

So why did Pieter Levels really build Hotelist?

Pieter Levels built Hotelist because years of heavy travel convinced him that choosing a good hotel still involves far too much guesswork, and AI finally made his preferred solution practical for one developer.

The origin is unusually easy to trace. During the Lex Fridman podcast, Levels was already complaining about inconsistent hotel quality while Fridman described unreliable Airbnbs, misleading hotel photos and rooms without proper desks. They explicitly discussed building a Nomad List-style product with much richer hotel information.

Hotelist now turns that conversation into software.

Levels also has a sharper complaint these days. He distrusts review systems where almost every property appears excellent, and he dislikes commercial programmes that can affect which hotels receive visibility. Booking.com's own disclosures and European competition investigations give the second part of that argument real weight, even if some of Levels' claims about review deletion are broader than we can independently prove.

AI solved another piece of the puzzle. Hotelist can currently process ratings, outside traveler experiences and photographs across roughly 90,500 hotels, then turn that material into searchable attributes and a single score. Doing that manually would have required a completely different organization.

The recent Airbnb review dispute clearly made Levels louder about Hotelist. It did not create the idea. We can trace the product back to an older and much more persistent frustration: he travels a lot, repeatedly found hotel information inadequate, and wanted a better system for himself.

The conclusion is pretty clear. Hotelist exists because Levels thinks the travel industry's weakest point is no longer finding a room to book. It is figuring out which room deserves to be booked in the first place.