Airbnb

Almost every impact metric is Airbnb's own, and denominated in inputs

Listings removed, tools launched, toggle users, tickets deflected, percentage of hosts who are individuals. What is not published: the host cancellation rate, AirCover claim approval and denial rates, mismatch incidence, chore incidence, refund-approval rates, or the accuracy of the anti-party model. Negative finding Inputs are what a company can control; outcomes are what a customer experiences. The score is capped by the absence of the second.

Half the platform is curated. The other half is the one with the problems

Guest Favorites is a genuine information gain and a genuine concession: by building it, Airbnb acknowledged that a filtered subset performs materially better than the whole. Analytical inference Every finding on this page about variance, remedy and support concentrates in the uncurated half, and a value score has to reflect the product a customer might actually be sold rather than only its best tranche.

The support layer is being automated on a stated cost rationale

Roughly 30% of English-language tickets resolved without a human in late 2025, about 45% by mid-2026, against a cost per booking falling around 10% a year, and an AI trained across "nearly 100 policies." Company statements Airbnb says service quality is improving and that its customer-service NPS is the strongest since the pandemic; the figure is unquantified and self-defined. Automation may well be better. It is being deployed at the precise layer where this page finds the customer most exposed.

The hotel comparisons Airbnb publishes are curated

Airbnb's own like-for-like figures — a one-bedroom at $114 against a hotel room at $148 in early 2024 — include the Airbnb cleaning fee, which is fair, but compare a one-bedroom home against all hotel rooms. An independent cross-check put the hotel figure at $140.16 on a series that excludes resort fees and includes China, and flagged the whole comparison as apples-to-oranges. Trade analysis caveat The comparison is not dishonest; it is selected. Independent all-in work reaches a different answer for the short solo trip.

The welfare literature is a decade old

The $41-per-night consumer surplus estimate rests on 2011–2014 data — before professionalisation, before the cleaning-fee norm, before the fee ceiling drifted, and before the derived average nightly price rose about 47%. Academic It is the strongest independent evidence of value in this file and it describes a version of the platform that no longer exists. No comparable post-2020 estimate was located.

Regulation removed value the customer used to receive

New York's Local Law 18 cut short-term-rental listings from roughly 22,000 to about 2,300; Spain ordered 65,935 listings removed in 2025; Japan's 2018 law cut listings from about 62,000 to 13,800 with mass cancellations. Regulatory record Airbnb handled the Japanese shock generously — a $10 million fund with full refunds and a coupon of at least 100% — but the customer-facing result across these markets is less inventory, higher taxes and, episodically, a cancelled trip. Not Airbnb's fault; still a reduction in delivered value.

Trust & Transparency — 10 / 15

Proportionally the heaviest dimension on this page, and the only one where Airbnb's own conduct — rather than its hosts', its history's or its regulators' — produces most of the harm. Five mechanisms are scored. The register gap in AirCover (3): protections marketed as guarantees and operated as a discretionary, evidence-gated, 72-hour programme whose remedy scales to "evidence strength." Arbitration (1): a class waiver and mandatory individual arbitration in the US — disclosed, with a 30-day opt-out and a 2021 carve-out for assault claims, but a material gap between the contract a customer thinks they hold and the one they can enforce. Terms of Service

The 21 April 2025 pairing (2): a headline transparency reform and a customer-adverse reduction in cleaning-fee refundability took effect the same day, and only one was announced. Scored here, once, and nowhere else — the entitlement reduction and the communication failure are the same action. Opacity where it is load-bearing (3): no published delisting threshold, no chore incidence, no AirCover claim-approval rate, no host cancellation rate, no anti-party accuracy metric, no ranking weights. Framing asymmetries (1): "90% of hosts are individuals" and "almost 40% of listings charge no cleaning fee" are both accurate and both chosen denominators. Technically true rhetoric is scored lightly, because it is technically true.

Five points withheld, and they are substantial. Airbnb publishes its ranking factors in unusual detail for an online travel agency, publishes its cancellation terms in full, published the camera ban and the checkout guidance in its own newsroom, publishes the DSA appeal numbers that this page uses against it, and fixed price display outright. The methodology counts disclosure and corrective action as mitigating evidence, and here it mitigates a great deal. Also struck on re-audit: the compressed rating scale, which is a real information problem but one Airbnb has answered with published badge thresholds, a bottom-10% signal and a lowest-rated-review sort. That is corrective disclosure, and it is not charged.

Customer Restriction — 8 / 20

Four mechanisms survive, and the dimension is capped hard by the thing it principally measures. The enforcement tail (3): during a pending-removal window access is limited, listings are hidden, payouts are paused and reservations with check-in inside that window are automatically cancelled, with guests not told why; an approved appeal restores access but does not reinstate the cancelled booking. In 2024, 41.2% of appealed restrictions were reversed. Screening that can deny access (2): tens of thousands of bookings blocked per US holiday period with no published accuracy metric, no false-positive rate and no documented appeal route for a blocked booking. Platform policy · regulator-mandated data

The fee-plus-labour structure (2): policy permits a guest to pay a distinct cleaning charge and be assigned part of the turnover, with no categorical prohibition and no published incidence. Association-based restriction (1): access denied on inferred association rather than the customer's own conduct, with documented false-positive harm. The cap: the anchor for the 12–15 band requires switching to carry significant friction, and on Airbnb it carries none at all — no subscription, no loyalty status, no accumulated standing, no data moat, no exit process. A dissatisfied customer books a hotel.

Twelve points withheld. Struck on re-audit: ordinary house rules and checkout instructions, which are intrinsic to sleeping in someone else's home, disclosed before booking and not scoreable as hostility; the 72-hour reporting window and evidence gate, which are the same architecture already charged under Trust and carry genuine operational necessity; and identity verification, whose access consequences are real but whose collection is proportionate. Security deposits are prohibited for most hosts, off-platform fees are banned, cancellation has loosened since 2018, and Lock-In, Exit Resistance and Loyalty Penalty were each tested and found not supported.

Revenue Extraction — 6 / 25

The dimension measures how aggressively Airbnb increases the customer value Airbnb captures. Applied literally, almost nothing qualifies. The cross-currency surcharge (3) is the one clean instance: from around April 2024 Airbnb added roughly two points for paying in a currency other than the listing's, transferring an exchange-rate cost it had previously absorbed onto the guest. Existence verified; date trade-reported The non-currency fee-ceiling drift (1) is the remainder: Help-Centre language moved from "typically under 14.2%" to a published band whose top is largely the surcharge itself. The historical baseline is trade-quoted rather than archived, and the flat take rate shows the drift did not change what Airbnb captures per booking. The platform's slice of the cleaning fee (2) is the third: on split-fee listings the guest service fee of 14.1%–16.5% is calculated on the booking subtotal, and the subtotal includes fees the host adds — of which the cleaning fee is one. Help Centre

Cleaning fees are scored here at Airbnb's own slice, and at nothing more. The fee itself is host revenue and is not charged to the platform. But on a split-fee listing Airbnb calculates its guest service fee on the booking subtotal, and that subtotal includes host-added fees — so Airbnb does earn roughly fourteen to sixteen cents on every dollar of cleaning fee, and earns more when the fee is larger. On a US cleaning fee in the observed $96–$175 range that is on the order of $14 to $28 per booking, for no additional platform service. The company that tells hosts to "aim to use the cleaning fee to cover the expense of cleaning – not to make additional money" takes a percentage of it either way. The 47% rise in the derived average nightly price stays excluded for the reason it always was — five and a half million independent sellers set it, and the platform's take rate did not move.

Nineteen points withheld, and the cleaning-fee finding is capped for four reasons. The rate is not elevated on the fee — it is the same percentage applied to the whole subtotal, which is ordinary marketplace commission on what the customer actually spends. It is fully disclosed inside the total price shown before booking. It applies to split-fee listings only: on the growing host-only-fee set there is no separate guest fee at all. And the flat take rate shows the component never expanded Airbnb's overall capture rate. Also struck: the 21 April 2025 refundability change, which transfers value to the host under the host's cancellation policy and is scored once under Trust; and the 2026 services layer — luggage storage, airport pickup, groceries — which is new value sold for money, not previously included value removed and resold. The derived take rate did not rise: 12.7% in 2019, 13.4% in 2025. There is no subscription, no paywall, no tier ladder, no advertising, no sponsored listings, no loyalty pricing and no personalised pricing anywhere in this file. For two of the three customer segments examined, Airbnb is materially cheaper than the alternative.

Behavioral Manipulation — 0 / 25

Tested against the literal anchor and scored at zero. The dimension measures the commercial use of attention, uncertainty, scarcity, urgency, psychologically salient signals and engagement mechanics to increase monetization — and requires "a meaningful customer cost or commercial manipulation component." The previous derivation rested almost entirely on the compressed review scale. On re-audit that does not qualify: rating compression is an emergent property of a reciprocal review system, documented by academics rather than designed by the company, and nothing in the research file establishes that Airbnb engineered or maintains it to increase bookings. Negative finding

The evidence runs the other way. Airbnb's one documented intervention in the review mechanism — the July 2014 simultaneous reveal — cut retaliatory one-star reviews by about two-thirds and had precisely estimated null effects on demand: a change made because the evidence said the system was distorted, with no commercial gain attached. Guest Favorites, the Top 1/5/10% labels, the bottom-10% signal and the lowest-rated-review sort all add differentiation back to a compressed scale. What remains is an information-quality problem, and information quality is a Trust question, not a manipulation one. It is discussed at length in this investigation and charged in neither dimension, because the corrective disclosure answers it.

All twenty-five points withheld. Historical drip pricing — the textbook mechanism, and the largest behavioural finding available in the file — was corrected by the December 2022 toggle, ranking by total price from November 2022 and the April 2025 global default, and the fixed-under-pressure principle credits it in full. The "guarantee" branding is an expectation gap and is charged under Trust. Beyond those, the file establishes no engineered countdown, no manufactured deadline, no variable-reward loop, no notification machinery, no gamified tier ladder, no scarcity cue, no personalised pricing and no sponsored placement. Search ranking optimises for booking probability — the peer-reviewed record says the model "already understood cheaper was better," and a 2025 objective explicitly reduced bookings likely to need customer support. A dimension worth a quarter of the total scale scores nothing, and that is the correct answer rather than an embarrassing one.

Information & Privacy — 3 / 15

Airbnb collects a great deal, monetises almost none of it, and has a strong legitimate reason for most of what it holds — a business in which a stranger sleeps inside someone's home. Two mechanisms survive. Relational inference (2): Airbnb confirmed in 2023 that it may restrict accounts of people "likely to travel with" a removed user, identified through shared payment cards or shared reservations. That is profiling and inference reaching beyond the customer's own transaction, applied with documented false-positive harm and no published accuracy. Retention persistence (1): ID images held around three years in many regions and selfies for the life of the account. Platform policy

Struck on re-audit: identity verification itself, which is proportionate to the transaction and is not scoreable merely because the data is sensitive; background checks, which are a legitimate marketplace safety mechanism whose consequences — wrongful or over-broad removals — are charged under Customer Restriction rather than here; and automated screening, because algorithmic decision-making is not by itself a privacy harm. The methodology is explicit that "the mere creation of useful customer data" is not hostility and that "consequence, consent, control, persistence, and use matter." Of those five prongs, only persistence and inference are met.

The camera ban runs the other way and is credited, not discounted. Indoor cameras were prohibited outright in April 2024, including previously disclosed and permitted ones — a categorical guest-side privacy gain stricter than hotel common-area norms, and a rare instance of a platform removing a capability from the side that pays it. Twelve points withheld. No advertising business, no sponsored listings, no behavioural ad targeting, no cross-service matching, no data brokerage and no third-party sharing beyond regulators and legal process is established anywhere in this file. Hosts never see a guest's ID.

How 27 sits on this site's scale

CHI 27 places Airbnb just below the Zoom / Rolex / Audemars Piguet cluster at 28, above Lamborghini's 18, and well below Canva at 38, Hermès at 48, Ticketmaster at 63 and Instagram at 70.

That is a materially lower number than this investigation first produced, and it is what a literal application of the methodology returns. Strip out what Airbnb's hosts do rather than Airbnb, what Airbnb has already corrected, what is intrinsic to renting a stranger's home, and what was being charged twice under different pattern names, and the remaining list is short: a currency surcharge it used to absorb, a percentage of every cleaning fee it tells hosts not to profit from, protection marketed harder than it is delivered, one adverse change made quietly, a set of unpublished metrics at exactly the points where a customer in trouble needs them, an enforcement layer that reverses a large share of what it is asked to reconsider, and a fee-plus-labour structure it permits. The finding did not weaken. The scoring got more honest about which company the harm belongs to.

Is this conduct already captured elsewhere?

It was, in one place worth six points: the remedy architecture was charged in three dimensions on the strength of three different pattern labels attaching to a single mechanism. It is now charged once, under Trust & Transparency, where the expectation gap lives. Both duplicate charges were struck.

Are we scoring a problem the company already fixed?

Yes, for seven points. Drip pricing is corrected, and under the fixed-under-pressure principle it is credited in full and documented as history. The regulatory timing appears in the reform ledger, where it belongs, and is not used to deny credit. The disclosure half of the cleaning-fee paradox is corrected on the same basis.

Are we scoring the platform for its sellers?

The largest theme in the audit, at eight points net. The nightly rate is not Airbnb's revenue and does not score against it. The cleaning fee is host revenue too — but Airbnb calculates its own fee on a subtotal that includes it, so the platform's slice of it does score. What Airbnb designs and earns on is charged; what its hosts charge and keep is not.

Have we subtracted the legitimate value?

Four points came off for operational necessity alone. Identity verification buys fraud and safety protection, background checks and anti-party screening buy property and neighbour protection, house rules buy predictable stays, and a reporting window buys usable evidence. Each burden was scored net of what it delivers, not gross.

Are we treating self-selected data as prevalence?

No. Appeal reversals are not an error rate, complaint-tweet corpora are not incidence, and press-reversal cases are not a resolution rate. Each is presented as what it is. Two points came off here, and the entire Publicity-Gated Remedy finding is labelled a mechanism rather than a rate.

Would it survive a reviewer who likes Airbnb?

That was the test applied to every paragraph, and on re-audit it removed more than half of the opening score. Each of the five easiest accusations — rising take rate, hidden fees, secret ranking manipulation, guest surveillance, worsening cancellation terms — is contradicted by the evidence and is refused on the page rather than softened. What remains is what a defender has to answer.

Deceptive Simplicity · Supported historically, partial currently

The legible proposition — a free guarantee, a 24-hour safety line, a platform that has your back — is materially simpler than the operative machinery, which is a 72-hour reporting window, an evidence requirement that can include the host's own confirmation, a remedy scaled to "evidence strength," and a Terms of Service disclaiming party status. The same gap appears in the framing asymmetries: 90% of hosts against 38% of listings, and 40% of listings with no cleaning fee against 89% of US listings with one.

Scored under Trust & Transparency, inside the same 10 of 15. Not scored again under Behavioral Manipulation, which scores nothing at all.

NOT SUPPORTED — tested, and rejected

Surveillance Creep, contradicted on the guest side by the categorical indoor-camera ban. Lock-In and Exit Resistance: no subscription, no status, no data moat, nothing to escape. Loyalty Penalty: no programme exists to penalise, and no evidence of differential pricing for repeat guests. Preference Amnesia: the total-price toggle persisted and then became the default — the opposite of the pattern. Algorithmic Replacement: ranking optimises booking probability with quality and support-risk terms, and no sponsored placement has launched. None scored

SUPPORTED — and scored

That the platform controls the transaction and disclaims party status to it. That its guarantees are discretionary, time-boxed and evidence-gated. That the all-in price inverted against hotels for the short solo or couple stay. That the guest fee ceiling drifted upward and an FX cost was shifted to guests. That cleaning fees are near-universal in the US and can coexist with assigned checkout labour within policy. That professional operators run the majority of listings in most large markets. That enforcement is opaque at first instance and that a substantial share of appealed decisions is reversed. That a headline reform and an unannounced adverse change took effect on the same day.

WOULD REDUCE THE FINDINGS

None of these requires Airbnb to abandon discretion, underwrite every stay, or become a hotel chain.

— Publish AirCover claim outcomes: how many reservation-issue reports are made, how many are approved, at what median remedy.
— Publish the host cancellation rate, not only the year-on-year delta.
— Publish an accuracy or false-positive measure for the anti-party model and background-check screening, and document a single clear appeal path for a blocked booking.
— Extend the 72-hour reporting window, or state plainly that late reports are accepted where the delay is reasonable, in the policy rather than as an aside.
— Prohibit specific checkout tasks outright — the same three the company already calls unreasonable — and publish how many listings were removed for chore complaints.
— Announce customer-adverse policy changes in the newsroom, with the same visibility as customer-positive ones. The 21 April 2025 exhibit would not exist if this had been done.
— Publish the numeric quality threshold at which a listing is warned or removed, ending the folklore.

WOULD STRENGTHEN THEM

Any of these would move the score materially upward.

— Sponsored listings launching. On record as an executive ambition since 2023 and modelled by analysts as a billion-dollar-plus opportunity. Paid placement inside a ranking system that currently optimises for booking probability would reopen Algorithmic Replacement immediately.
— A rising take rate. The flat 13% is doing a great deal of work on this page; if it moves, several findings change with it.
— Evidence that support automation is degrading outcomes rather than only cost — a rising unresolved-case rate, or a falling reversal rate paired with rising complaint volume.
— A further narrowing of refund entitlements without announcement. One instance is an exhibit; a pattern would be a finding.
— Any documented instance of a guest being penalised for a public complaint, which would convert Publicity-Gated Remedy from a mechanism into something much worse.
— US enforcement data showing appeal reversal rates comparable to the EU figures, which would remove the last defence that this is an EU-scope artefact.