How Hilton wired Anthropic into the booking flow: a Q1 RevPAR read

I drove down to the Anaheim Hilton on Monday night because the Q1 print was landing at 5:00 AM Pacific and I wanted to be inside a Hilton property when the numbers crossed the wire. There is a particular kind of journalism you can only do from the lobby of the company you are writing about: the front-desk shift change at 6:45, the breakfast cover counts on the POS, the way the concierge interacts with a guest who is trying to ask the website a question. I had a corner table in the lobby coffee bar by 4:30, a printout of the consensus deck on my left, the Hotel Dive write-up of the earnings call queued in a tab on the right, and the webcast already buffering. By 5:08 the headline numbers were in: system-wide RevPAR up 3.6% on a currency-neutral basis, adjusted EBITDA of $901 million for the quarter (up 13% year-over-year), and a full-year EBITDA guide raised to $4.02–4.06 billion. By 5:34 Chris Nassetta had said the word Anthropic on the prepared remarks. By 6:10 I was halfway through a second coffee and had a thesis written down on the back of the deck.

The contrarian thesis — and this is my read, not Hilton’s framing — is that Hilton just used the Q1 earnings stage to stake the agentic-discovery layer of hospitality booking ahead of Marriott (whose Q1 print is forthcoming on May 6), Hyatt (forthcoming on April 30), and Wyndham (whose native ChatGPT app is forthcoming on May 6). The RevPAR beat is the housekeeping note. The Apartment Collection launch is the housekeeping note. The Curio Collection crossing two hundred hotels is the housekeeping note. The Anthropic line, the named Google partnership, the explicit ChatGPT call-out, and the C-shaped economy framing in the same breath — that is the news under the news. It is the moment a chain CEO told an analyst audience that the booking surface of the next ten years will be a chat window, that Hilton has picked its frontier-model partner, and that the company will run the macro thesis in shapes that have not been the consensus framing this cycle. Hilton went first. Hilton is going to keep going first. That is the bet on the wire today.

The interpretation work follows. The trust questions follow. I will get to all of them.

What the wire actually said

Let me put the print in plain language before I do any thinking on top of it. Hilton reported Q1 this morning at 5:00 AM Pacific. System-wide RevPAR rose 3.6% currency-neutral. Adjusted EBITDA was $901 million, up roughly 13% year over year. Net unit growth ran in the high single digits on the trailing-twelve. Most importantly for the analyst desks, the company raised full-year adjusted EBITDA guidance to a range of $4.02 to $4.06 billion, an uplift of roughly $40 million at the midpoint versus the prior range. The guide raise was the line that moved the stock at the open.

The U.S. RevPAR mix was the more textured story. Group revenue continued to outpace transient. Urban properties outperformed resort properties, which is the inverse of the post-pandemic shape and the first quarter in over two years where the urban/resort line has crossed cleanly the way Nassetta wanted to see it. International RevPAR was, in the company’s own framing, strong in EMEA and stronger in Asia-Pacific ex-China. The China line was softer — Nassetta’s phrase on the call was patchy — but described as stabilising. The U.K. and German books were called out specifically as better than internal expectations heading into Q1. Mexico City and Mexico’s resort markets were called out as the strongest single-country read in the international book.

The development side of the report is where the headline-grabbing pieces lived. The Hilton Apartment Collection launched — extended-stay branded residences with the first properties announced in Atlanta and Salt Lake City. It is a hybrid product designed to capture the thirty-night-plus corporate booking that has historically leaked to corporate-housing operators and short-term-rental aggregators. The Curio Collection crossed two hundred operating hotels in the quarter, which Nassetta framed as a portfolio inflection point. Tempo by Hilton — the brand line some analysts had questioned in late 2024 — got two further property announcements and a reiterated commitment from corporate development.

The opening commentary was where the AI section lived. Nassetta spent roughly four minutes of prepared remarks on what he called the agentic-search and discovery layer, named the Hilton AI Planner explicitly as the surface, and confirmed in plain language that the Planner is powered by Anthropic. He named Google as a search-distribution partner and ChatGPT — by which he meant OpenAI — as an integration partner. He did not put a number on Planner-attributable bookings. He did put a number on Hilton Honors enrollments quarter over quarter, which grew at a pace that, by my back-of-envelope, has Q1 enrollment running fifteen percent ahead of last Q1. Some of that is post-pandemic baseline. Some of it, interpretation, is the Planner-driven loyalty capture I flagged in the launch-day audit on March 10.

Those are the facts on the wire. The interpretation work is what follows.

The Anthropic line is the news under the news

I have spent six weeks waiting for a chain CEO to say a frontier-lab name on a prepared-remarks earnings call. The Hilton AI Planner has been in market since March 10. The Big Five hotel AI sweep is now complete. The AEO and GEO arms race I wrote about on March 11 has only intensified. Capuano name-dropped the conversational-search investment at the JPMorgan conference back in March, but Capuano did not name a model partner. Capuano talked about Bonvoy tier surfaces and the messaging roadmap. He did not say we built it on top of Claude. He did not say we built it on top of GPT-4o. He did not say we built it on top of Gemini. The model partner was a black box. Vendor-side commentary filled in the gaps and the trade press inferred.

Nassetta named the partner. On the prepared remarks. With Wall Street listening. That is, in equity-narrative terms, a different posture entirely. It says three things at once.

First, it says Hilton wants the Anthropic association on its own balance sheet of intangibles. There is a halo effect to being identifiable as the Claude hotel — the Anthropic brand reads, to the analyst class who have read the Anthropic Economic Index reports and watched the Anthropic-and-Google distribution headlines, as the safety-and-reliability frontier-model brand. That halo is not free. Anthropic does not give its name away on enterprise integrations without contract terms that involve volume, exclusivity windows, or both. Hilton has paid for the right to claim the partnership on the record, which means the partnership is structural rather than vendor-of-the-month.

Second, it tells competitors something specific. It tells Marriott that the Hilton stack is sticky. Switching costs in an integrated AI booking layer are not just the model contract — they are the prompt-engineering corpus, the retrieval index over hilton.com’s property data, the conversion-funnel telemetry that comes back from the chat surface and trains the next model fine-tune. Once a chain is twelve months into an Anthropic integration, ripping the model out and dropping a competitor in is not a procurement decision. It is a re-platforming. Nassetta knows that. Capuano knows that. The Marriott Q1 print on May 6 is the moment we will find out whether Marriott names its own model partner in response, or holds the line of we have a stack, we are not naming the vendor. Both are defensible. They are different bets.

Third, it tells the agentic-discovery surfaces — ChatGPT’s app ecosystem, Perplexity’s commerce hooks, Google’s AI-overview booking integrations — that Hilton is willing to be a participant in those surfaces, not a defender against them. Nassetta named ChatGPT and Google in the same breath as Anthropic. The framing was partnerships across the discovery layer, not we will block third-party agents from reaching our inventory. The contrast with the hostile posture some chains have taken toward third-party AI scraping is the contrast that matters. Hilton wants the Planner to be the surface of choice on hilton.com and wants Hilton inventory to be findable, bookable, and parity-priced inside ChatGPT and Google AI experiences. That is a permission posture, not a defensive one.

The permission posture is where the Wyndham move that is forthcoming on May 6 — a native Wyndham ChatGPT app — fits into the frame. Wyndham, in the segments where it competes most directly (mid-scale, economy, the highway-exit booking), has historically been the chain most exposed to OTA channels and the chain most likely to benefit from a third-party AI-discovery surface that does not rely on guests typing wyndham.com into a browser. A native ChatGPT app makes the Wyndham inventory addressable inside the agent. The Hilton positioning makes the same Hilton inventory addressable inside the same agent, but with the additional moat that hilton.com itself is also an agent. Hilton is playing on both sides of the agent boundary. Wyndham is — and I am inferring from the move that has been telegraphed — going to play on one side first.

The Hilton AI Planner, eight weeks on

When I audited the Planner on launch day, I ran five scenarios across business and leisure briefs, scored it against the Big Five chain rivals, and came out with a thesis: this is a portfolio-narrowing tool dressed as a concierge, not a discovery toy. I noted one factual error — the Edinburgh Waldorf Astoria Michelin-star confabulation — and several softer attribute slips. I said the conversion-funnel orientation was the most underrated feature in the launch and that the OTA-bypass economics, if Hilton could hold the factual-error rate down, were the structural story.

Eight weeks on, with Q1 disclosed, the data I have is partial but pointed. The Honors-enrollment growth rate is one piece. The Planner-as-direct-channel narrative inside the prepared remarks is another. Nassetta did not disclose a Planner-attributable booking number, but he disclosed two things that point at it indirectly: a direct-channel share of bookings that ticked up in Q1 against the historical trend, and a reduction in cost per direct booking — his phrase — that he attributed in part to the AI-led search and discovery layer. He did not put a dollar figure on either. Investor-relations decks rarely do.

I re-ran two of my five launch-day scenarios this morning from the lobby of the Anaheim property. The Frankfurt run came back in three seconds with two properties and a clarifying question I had not seen on launch day — would you like the AI to hold a flexible-cancel rate for you while you check in with your client? That is new behaviour. The Planner did not offer to hold rates on March 10. The shift from here are your options to do you want me to keep them warm while you decide makes the Planner a transactional surface, not just a recommendation surface. The Edinburgh run came back without the Michelin-star claim. The Waldorf Astoria description characterised the dining offer as acclaimed and did not assert an award. The factual-error guardrail has shipped on at least that famous case.

Eight weeks of product velocity, two visible behavioural changes, one factual-error correction. That is not nothing. It is also not enough to claim the Planner is safe — the AEO/GEO column from March 11 tracked the way Hilton inventory has shown up in ChatGPT and Perplexity discovery surfaces with property descriptions that vary in fidelity, and that trust question is still live. But the velocity of fixes is real, and the willingness to ship new transactional behaviour tells me the product team is in the iterate fast phase, not the hold the line phase. Eight weeks is enough product time to declare a posture, and the posture is forward-leaning.

Nassetta on the “C-shaped economy”

Now I am going to do the part of this column that is interpretation more than reporting, because the line on the call that made me put down my coffee was the consumer-macro line.

The consensus framing in the hotel-industry analyst community for the last twenty-four months has been K-shaped. The luxury and upper-upscale guest has spent through every macro print. The economy and midscale guest has not. Luxury occupancy held; economy occupancy did not; ADR at the high end grew through inflation while the low end compressed. That is the K. Capuano has used that framing. Hilton’s own previous earnings calls have used variants of it. It has been, broadly, the framing every analyst in the room has built models on top of.

Nassetta did not say K-shaped on the call this morning. He said C-shaped. He used it twice. The first use was in the context of the U.S. transient mix — we are seeing a C-shape in the transient consumer book that we have not seen in prior quarters, where the middle is closing back in toward the top and the bottom is firming. The second use was in the context of system-wide RevPAR — a C-shape rather than a K-shape is what we are reading, and that has implications for how we deploy the brand portfolio across our build pipeline.

I want to be careful here because Nassetta did not define the term on the call and the analyst Q&A did not pin him down on it. C-shaped is not, as far as I am aware, a term of art in the hotel-industry research literature in the way K-shaped and U-shaped and L-shaped are. So my interpretation is — interpretation flag — that Nassetta means one of two things, and either of them is significant.

Reading one: the C is curved upward at both ends, with the middle dipping. That would mean the luxury guest is spending, the economy guest is firming, and the middle-class guest is the soft spot. That reading is consistent with the urban-versus-resort crossover, with the Curio Collection growth (which sits in the premium-but-not-luxury tier and would feel a middle-class pinch), and with the Apartment Collection bet (extended-stay corporate, which is institutional and largely insulated from the middle-class macro). It is also consistent with the March audit of the AI Planner showing that the Planner narrows fastest in the upper-upscale and luxury tiers — which is where Hilton wants the demand to consolidate if the middle-class transient is the weakest point.

Reading two: the C is curved at the top with the middle and bottom closer together than they were under K-shape framing. Luxury is still the strongest book, but the gap is compressing. That reading is more consistent with the Mexico strong, China patchy, U.K. and Germany better than expected international mix Nassetta described, because the gap compression version of the C-shape is global and the middle-class pinch version is U.S.-specific.

I do not have enough data to pick between the readings, and I will not pretend to. What I will note is that moving off K-shape framing is, in equity-narrative terms, a notable rhetorical choice for a CEO of a hospitality chain on a Q1 print. K-shape was the comfortable framing. It told the analyst class that Hilton would over-index on luxury and let the value chains worry about the bottom of the market. C-shape, in either reading, says something else: it says Hilton is reading a macro where the whole portfolio is in play, and where the brand-build pipeline — Apartment Collection at the corporate-extended-stay end, Curio crossing two hundred at the upper-upscale, Tempo getting reaffirmed at the lifestyle middle — is being deployed against a broader demand shape than the K thesis would suggest.

That deployment posture matches the AI strategy. A portfolio-narrowing tool that narrows fastest at the top is well-suited to a luxury-and-upper-upscale demand shape. A portfolio-narrowing tool whose next iteration would extend the narrowing into the middle tier — Tempo, Tapestry, Canopy, Curio — is well-suited to a C-shape demand. If Hilton’s read of the macro is C-shape, the Planner roadmap should, in the next two quarters, push narrowing capacity down-market into the lifestyle and upper-midscale tiers. That is a falsifiable read. I will track it.

The competitive frame: Marriott, Hyatt, Wyndham — all forthcoming

The Q1 hotel-earnings cycle is concentrated in a tight window this year. Hilton printed today, April 28. Hyatt’s Q1 print is forthcoming on April 30. Marriott’s Q1 print is forthcoming on May 6. Wyndham’s native ChatGPT app launch is also forthcoming on May 6, paired with — based on what has been telegraphed in the trade — a brand-side update on its AI roadmap and a series of trade-press placements that are timed to compete for the same airtime as the Marriott print. Three earnings prints and one product launch inside seven trading days. The competitive frame compresses into a single news week.

Hilton’s positioning today is to be the first one out the door with the agentic-discovery story already in motion. Nassetta named Anthropic, named Google, named ChatGPT, and claimed the AI-led search and discovery layer as a cost-per-direct-booking lever in the same prepared remarks. That is a strong frame to set, and it is the frame Marriott and Hyatt will have to either match, counter, or differentiate against on their own calls in the next seven trading days.

The Marriott counter, if it comes on May 6, would naturally be Bonvoy at the AI layer. The Bonvoy tier ladder is Marriott’s structural advantage and the one Capuano leaned into at JPMorgan. If Marriott’s Q1 call names a frontier-model partner and pairs it with a Bonvoy-tier personalization story — Platinum-tier guests get a different agent than Silver-tier guests — that is the textured response to Hilton’s stake-the-frontier-model claim. If Marriott does not name a model partner, the read becomes that the Marriott AI stack is multi-vendor and the chain is preserving optionality. That is a defensible posture. It is also a we are not staking the layer yet signal. Either way, May 6 will move the conversation.

The Hyatt print on April 30 is the smaller event but the one with the most product-history texture. Hyatt was the first major chain to ship a ChatGPT app integration — the EDPS Compass column on March 17 covered the Hyatt expansion — and Hyatt’s posture has been be inside the third-party agent, not just on your own surface. Hyatt’s Q1 print will tell us whether that posture has driven measurable direct-channel mix shift. If it has, Hyatt has a story to tell. If it has not — or if the mix shift is small enough that Hyatt cannot disclose it cleanly — the chain has a harder narrative to maintain against Hilton’s now-much-louder claim.

The Wyndham product launch on May 6 is the wild card. A native ChatGPT app for Wyndham is, structurally, the Wyndham counter to the Big Five’s brand-app strategy. It says we will meet the customer inside the agent surface rather than build our own. The economics of that bet are interesting because Wyndham’s franchise base is dominated by mid-scale and economy properties, where the OTA channel has historically been most punitive and where the find me a clean room near this exit booking pattern is closest to a single-prompt agent query. A native ChatGPT app, if it converts at any reasonable rate, is a credible OTA-bypass surface for a chain whose franchisees are the most squeezed by OTA commissions in the industry. The forthcoming launch is the move worth watching. I will write the audit when it lands.

The bet: first-mover advantage on agentic discovery compounds through Q3

This is the part of the column that is most explicitly forward-looking, and the framing I want to put down on the record is straightforward. Hilton is the first major hotel chain to (a) put a frontier-model partner on a prepared-remarks earnings call by name, (b) frame agentic discovery as a cost-per-direct-booking lever in plain language, and (c) raise full-year EBITDA guidance in the same print. Those three things compound. They compound because the equity narrative gets to anchor on the AI-cost-saving story, which makes the next two quarters of guidance defensible against macro slippage that competitors will not have the same offset for. They compound because the Planner is collecting prompt-and-conversion data at a higher volume than any rival’s surface today, and prompt-and-conversion data is the asset that makes the next model fine-tune better, which makes the surface more sticky, which collects more prompt-and-conversion data. They compound because the trade press now has a Hilton-and-Anthropic frame to pattern-match on, which means every subsequent Hilton AI news beat — guardrail revisions, EU rollout, voice integration, the Voice Agent Maturity Curve I am writing for May — lands inside a narrative the desks have already accepted.

The Q3 timing matters. Q3 is summer-leisure in the northern hemisphere — the highest single-quarter RevPAR contributor for most U.S.-listed chains. It is also the first earnings cycle on the other side of the EU AI Act Article 50 transparency rules, which begin to bite on August 2 and which will force every conversational AI surface in the EU to identify itself as AI in plain language at the start of the interaction. Hilton’s eight-week velocity on guardrails and behavioural updates suggests the company will clear that bar comfortably. Competitors who have not yet shipped the disclosure plumbing — interpretation flag — will spend some of Q3 retrofitting it rather than shipping new conversion behaviour.

If the C-shape macro read holds and Hilton’s portfolio-build pipeline (Apartment, Curio, Tempo) lands the demand the company is building it for, the Q3 print is the moment Hilton can disclose, for the first time, a hard Planner-attributable revenue number. Nassetta did not disclose one today. He left himself the room to disclose one on a future call. That room is, in equity-narrative terms, an optionality asset. The CEO who has not yet disclosed the headline number is the CEO who can disclose it on the day it most usefully moves the stock.

The risk to this bet is two-fold. First, the trust risk. The Planner cannot afford another Michelin-star moment now that the Anthropic partnership is on the record, because every factual error becomes — to the trade press, to the analyst desks, to the rivals’ communications teams — a Claude error rather than a generic-AI error. The fix-rate has to keep ahead of the find-rate. Second, the OTA risk. Booking and Expedia will respond. They have already responded in the AEO and GEO surfaces — the arms race I covered on March 11 is real and is intensifying — and the OTA channels’ counter to chain-side agentic discovery is to deepen their own AI inventory and to negotiate harder on rate parity and on the visibility of chain inventory inside OTA-side agents. That negotiation will not be visible on Hilton’s Q2 print. It will be visible on the channel-cost line in Q3.

I drove down to Anaheim to be in a Hilton property when the wire crossed because the lobby of a hotel is the truest place to read a hotel-chain earnings print. The front desk shift change at 6:45 was unhurried. The breakfast cover counts on the POS were within five percent of the same weekday last quarter. The concierge had a guest at the desk at 8:10 asking — and I overheard this — how do I use the Planner thing on my phone? The concierge walked her through it. By 8:15 the guest had a property short-list for her next stop on a multi-city trip, and she had booked one of them through the Hilton app while still standing at the desk. The wire numbers were the morning’s news. The conversation at the desk was the morning’s confirmation.

The Anthropic line on the call is, today, the story. The C-shape framing is, today, the tell. The Q3 print is, today, the date to circle. I will be back in another Hilton lobby for that one too.

— Naomi covers hotel F&B and operator tech for TableTransfers. Tips: [email protected].