A Host Skipping the Approval Gate Helps the Market. The Platform Forcing Every Host to Skip It Doesn't.

Based on the research ofChen, Feng, Feng, Jiang, and Wu, "Platform Governance Through Consumer Screening: A Theoretical and Empirical Analysis," MIS Quarterly, 2026

Chen, Feng, Feng, Jiang, and Wu built a game-theoretic model of a decision every Airbnb host faces: keep the guest-approval step, where you personally vet every booking request, or turn it off and let anyone with a clean-enough profile book instantly, the mode Airbnb calls Instant Book. Validated against real Airbnb data, their model produces a genuinely counterintuitive result. A low-rated host who forgoes screening does not necessarily raise its price to capture the new convenience premium; it can actually lower price, because a rival reacts aggressively to the competitive threat. A high-rated host, meanwhile, gains almost nothing from dropping screening, since guests already trust it and convenience was never the binding constraint. But in markets with real competitive intensity, one provider's choice to forgo screening ripples outward and lifts the provider, the rival, and the platform together, raising total welfare. The reversal comes at the platform layer: when Airbnb itself engineers the screening default, mandating or nudging providers toward Instant Book through feature design rather than leaving it a provider's choice, it can raise its own profit while total welfare falls. Same lever. Opposite outcome, depending entirely on who is holding it.

Instant Book: from opt-in convenience to engineered default

Airbnb's own account of Instant Book traces the feature to 2014, built to let guests book a listing immediately, no message to the host, no waiting for approval. For its first two years it was one option among several, a convenience toggle a host could flip or leave alone. That changed in September 2016, when Airbnb released a review led by former ACLU Washington office director Laura Murphy on discrimination on the platform, following widespread reports of hosts rejecting guests based on race under the #AirbnbWhileBlack banner. One of the plan's concrete commitments was to grow Instant Book to one million listings by January 1, 2017, explicitly because a booking that never passes through a host's individual approval has no moment at which that host can discriminate. Airbnb hit the goal early, reporting one million Instant Book listings that December.

That 2016 push was not a host-by-host decision anymore. It was the platform choosing, market-wide, to shrink how much any individual host got to screen. Airbnb has kept building in that direction ever since. Its engineering team describes Instant Book today as "a key lever" the company actively promotes, and has shipped a machine-learning-driven Smart Instant Book Filter that surfaces Instant Book listings more prominently to guests who show intent to book quickly. Airbnb's own current figures, published on its newsroom site, put the scale of that push in view: more than 40 percent of all listings are now instantly bookable, roughly 70 percent of new listings turn it on from their very first day, and 60 percent of guest bookings are now secured through Instant Book. What began as a host's option has become the product's engineered center of gravity, which is exactly the shift the paper's welfare warning is about: an individual host toggling this off is one thing, and a platform steering the entire marketplace toward it through search placement and onboarding defaults is another.

The provider's choice versus the platform's design

The paper's mechanism explains why that distinction matters financially, not just ethically. When a single host forgoes screening, the effect runs through a rival's pricing response, and that response depends on local competitive intensity. Where competition is thick, one host's move to frictionless booking forces a rival to compete harder on price or service, and the surplus from that contest spreads across both hosts and the platform that intermediates them. The host's individual calculation, in other words, is disciplined by the fact that a rival is watching and will react.

A platform mandating the default removes that discipline. When Airbnb boosts Instant Book listings in search rank or pushes new hosts to enable it at signup, it is not responding to a rival's competitive move: it is setting the rules that determine how much any host, anywhere, gets to screen at all. The paper's finding is that this kind of platform-level engineering can raise Airbnb's own profit even while lowering total welfare, because the platform's revenue depends on conversion and booking volume, not on whether the resulting market allocation is efficient for hosts and guests. That is the paper's regulatory hook: individual provider screening decisions, filtered through competition, tend to self-correct toward outcomes that help the whole market; platform-level screening design does not carry that same discipline, and needs outside scrutiny instead.

When ride-hailing ran a version of the same experiment

Airbnb is not the only marketplace where a platform has directly engineered how much a provider gets to screen the other side. Ride-hailing offers an unusually clean natural experiment, because Uber ran the toggle in both directions inside a few years, in the same state, for the same reason: California's AB5 gig-worker law.

In December 2019, facing pressure to prove its drivers were more like independent contractors than employees, Uber gave California drivers something drivers elsewhere did not have: the ability to see a rider's destination and fare estimate before accepting a trip, effectively a screening tool over which rides to take. Consumer advocates immediately flagged the risk. Hana Creger of the Greenlining Institute warned in February 2020 that the change "could only exacerbate discrimination," pointing specifically to the possibility that drivers would decline trips into lower-income neighborhoods like East Oakland or Bayview-Hunters Point, areas already underserved by transit. Uber's counter was procedural, stating that rejecting a trip specifically to avoid a neighborhood violated its community guidelines and California law, but the guideline could not undo what the new screening capability itself made possible.

The platform then reversed a related piece of that same driver-autonomy push. By April 2021, Uber ended a companion feature that had let California drivers set their own price multiplier on trips, after what Forbes reported as a 117 percent jump in rider cancellations; Uber told reporters that 80 percent of riders offered a fare above the standard multiplier declined it and did not rebook. The platform had engineered more provider discretion into the product to win a legal argument, watched it degrade reliability for the other side of the market, and then reversed just the piece that was hurting volume, not necessarily the piece raising discrimination risk. That is the paper's warning in miniature: a feature-level design choice made for the platform's own strategic reasons, adjusted again for the platform's own reasons, with the welfare and fairness consequences arriving as a side effect either way.

Two freelance marketplaces, two structural defaults

The same fork shows up, more quietly, in how freelance marketplaces are built from the ground up. Fiverr's entire model runs on gigs a seller lists at a fixed scope and price; a buyer purchases directly, and the transaction proceeds without the seller individually approving that specific client, structurally closer to Instant Book than to a request-and-accept system. Upwork went the other way by design: a client posts a job, freelancers submit proposals, and the client screens and selects among them before any contract exists, the marketplace equivalent of keeping guest approval switched on for every booking. Neither company frames this as a per-provider toggle. It is baked into the product category each one built, which is precisely the platform-level design choice the paper distinguishes from an individual provider's discretion. Upwork's own comparison materials position its model as suited to higher-stakes, longer engagements where client-side screening earns its cost; Fiverr's position frictionless purchase as the point for smaller, well-defined tasks. Each platform picked a default for its entire market, and neither pretends that choice is a host-by-host decision the way Airbnb still technically allows Instant Book to be.

A host who stops screening is placing a bet the market can correct. A platform that stops screening for every host is placing a bet only the platform gets to collect on.

The paper's cleanest lesson is that "screening" is not one policy question but two, and they do not answer to the same logic. Ask whether a single provider should be free to forgo it, and competition mostly polices the answer, spreading gains to rivals and the platform along the way. Ask whether the platform should be allowed to engineer that choice into the product for everyone, through ranking, defaults, or an outright mandate, and competition no longer polices anything, because the platform sits above the competitors it is supposed to be disciplined by. Airbnb built Instant Book as an option in 2014 and has spent a decade since turning it into infrastructure. That arc, and Uber's parallel back-and-forth over driver discretion in California, are what the paper's call for regulating platform-level screening design is actually about.