AI hotel pricing falls under FTC authority, unlike airline pricing.

Anyone setting AI hotel pricing in the United States acquired a regulator last month. The Federal Trade Commission proposed an enforcement policy statement on personalized pricing on 19 August 2026, defining the practice as using consumer data such as browsing history, location, demographics or purchasing patterns to set an individual price based on estimated willingness to pay. The comment period, extended by seven days, closes on 25 September. Section 5 of the Federal Trade Commission Act, which is the authority the statement rests on, carves out air carriers and foreign air carriers by name.

This piece is for independent hoteliers reading two years of headlines about airline pricing algorithms and wondering whether any of it applies to them. It applies more directly to them than to the airlines, and the reason is jurisdictional rather than technological. This article replaces an earlier version on this page that took a trade opinion column as its starting point and did not establish where the enforcement risk actually sits.

The theory the trade press is arguing about has never been shown to happen

Start with what is not established, because the earlier version of this piece did not.

The proposition circulating in hospitality commentary is that an artificial intelligence travel assistant will read a traveler's calendar, flight confirmations and browsing behavior, infer how badly that person needs a room, and add a margin on top of the rate the hotel published, keeping the difference. No platform has been shown doing this to hotel inventory. No filing describes it, no regulator has alleged it, and the column proposing it offered no instance of it.

What has been documented is the nearest analogue, in aviation. Delta Air Lines has been using a pricing system built with the Israeli company Fetcherr, described on its own earnings call as covering a small share of domestic fares with an intention to expand. Delta has twice denied to Congress that it uses personal data to set individual fares, most recently in a letter from president Peter Carter to Representative Frank Pallone dated 9 September 2026, and states that its system works from aggregated market information. Pallone had written to eight airlines on 11 August. That exchange is the whole public record, and it is a record of denial rather than of discovery.

The distinction regulators are drawing is worth learning, because it decides everything downstream. Dynamic pricing responds to the market: inventory, demand, seasonality, the competitor down the road. Personalized pricing responds to the individual. A revenue management system that moves a rate because the city is full is the first. A system that moves the same rate for one shopper and not another is the second. Nothing in the first category is at risk here.

The exemption that protects airlines leaves AI hotel pricing exposed

Here is the part that has not appeared in the hotel trade coverage at all.

Section 5 of the Federal Trade Commission Act empowers the Commission to act against unfair or deceptive practices, and then lists the exceptions. Common carriers are excepted. So are air carriers and foreign air carriers subject to part A of subtitle VII of title 49. That is why the pressure on airline pricing has arrived as letters from a congressman rather than as an enforcement action, and why bills have periodically been introduced in Congress proposing to strike the air carrier words out of the section altogether. Removing the exemption requires an Act of Congress, which is a fair measure of how solid it is.

Hotels are not in that list. Neither are the online travel agencies, metasearch engines or booking platforms that sell hotel rooms. An accommodation business or a platform selling accommodation that sets prices from personal data sits inside Section 5 in the ordinary way.

The proposed statement asks for three disclosures where consumers reasonably expect prices to be static: that the price is personalized, the basis on which it was personalized, and the types of data used. The Commission says vague phrasing such as "specially selected" will not meet that standard. It also concedes the limit of its own reach, stating that it does not have legal authority to ban personalized pricing in all circumstances, which places the entire statement on disclosure rather than prohibition. Chairman Andrew Ferguson's framing on release was that businesses engaged in or considering personalized pricing are on notice.

Beneath the federal layer, states have moved faster and further. In its analysis of the proposed statement, the law firm Hunton Andrews Kurth records that Connecticut has amended its data privacy law to impose both disclosure duties and substantive bans on surveillance pricing for retailers and third-party delivery services, that New York has passed a bill prohibiting it, and that California's attorney general has opened an investigative sweep into businesses using personal data for algorithmic pricing under the state privacy act. Those accounts are a law firm's reading of the state measures rather than the statutes themselves, and we have not read the underlying texts. What is not in dispute is that none of those regimes carries the air carrier exception either.

Agoda shipped the data asymmetry the theory depends on, and disclosed none of the logic

The mechanism the theory needs is a platform that holds a rich behavioral picture of a traveler and controls what that traveler sees. That is not hypothetical, and a current example arrived on 15 September.

Agoda, which is part of Booking Holdings, began a limited public beta of an AI Assistant for selected users of its mobile app. The release describes conversational trip planning connected to live inventory of more than six million properties, over 130,000 flight routes and more than 300,000 activities. Conversational intelligence runs on OpenAI's GPT-5.6, and for more complex itineraries the assistant uses Google Gemini's Grounding with Google Maps to pull real-time ratings, photographs and opening hours. In Agoda's own words, the assistant retains the traveler's preferences and trip context as plans develop.

That last sentence is the data asymmetry stated by the platform itself. The assistant knows who the traveler is across sessions. The property does not, and cannot.

What the release does not contain is any account of how the assistant decides which properties to name. When a traveler asks for a relaxed place near the water that takes dogs, three or four properties come back rather than three hundred, and nothing published says whether that selection turns on price, commission, review score, the richness of the listing description or a partnership. Chief executive Omri Morgenshtern's quoted description of the product is about removing planning friction and says nothing about ranking. The gap is the finding, and it is the same gap four assistants produced when we tested them in Miami, where the property that appeared on one assistant's list was close to uninformative about the others.

The counter-case is that the merchant arrangement makes the markup hard to hide

The strongest argument against reading any of this as an imminent threat is structural, and it comes from the platforms' own mechanics.

When Google launched hotel booking inside AI Mode, it stated that the hotel or booking platform remains the merchant of record and handles the charge, the confirmation and customer service, a point covered when Google's booking agent launched with ten partners. Where the property is the merchant of record, the guest is charged the property's rate by the property, and a silent premium inserted by an assistant in between has nowhere obvious to sit. The same is broadly true of agency-model bookings, where the hotel sets and receives the rate.

Merchant-model bookings are different, because the platform takes the traveler's money and settles with the hotel separately, which is the arrangement behind the $10.1 billion of traveler cash Booking Holdings was holding at 30 June. Under that structure the gap between what the guest pays and what the hotel receives is already the platform's, by contract and in the open. Nobody needs an assistant to create it.

So the honest position is that the mechanism the column describes would require either a merchant arrangement where the spread is already disclosed, or an agency arrangement where the hotel would see the discrepancy on its own confirmations. Neither makes it impossible. Both make it visible to somebody.

What the record cannot settle

No published document shows a platform varying a hotel rate by a traveler's inferred urgency. The FTC's statement is proposed rather than final, its comment window closes in two days, and it addresses disclosure rather than the practice. Agoda has disclosed its models and its inventory and nothing about selection or pricing logic.

Some operators are responding by checking what their own rate looks like from a clean browser against a logged-in session on each platform they sell through, on the reasoning that a discrepancy they cannot see is one they cannot raise. Others treat the whole question as a distraction from the commission line they can already read on every invoice.

The one thing the record does establish is where any test case will land. An airline that sets a price from a passenger's personal data answers to the Department of Transportation and to congressional letters. A hotel or a hotel-selling platform that does the same answers to the Federal Trade Commission under Section 5, to Connecticut, to New York and to the California attorney general, and the comment period on how that will be enforced closes on Friday.