Hilton's AI Planner, the Anthropic Bet, and What Q1 2025 Says About the Difference Between 'Deploying AI' and 'Owning the AI Agent'

I was sitting on the end of a king bed at a Hilton property in midtown this morning, laptop balanced on a luggage rack, watching the Q1 2025 earnings PDF take its sweet time to download over hotel Wi-Fi. The irony was not lost on me. The release I was waiting for is, among other things, a document about Hilton’s bet that the future of hotel demand runs through software the company itself controls — and I was reading it on Wi-Fi that, like the rest of the lodging stack, the brand technically does not own.

When the file finally rendered, I did the thing I always do first on a hotel print: skipped the cover page, skipped the safe-harbor paragraph, and went straight to the CEO quote. Chris Nassetta, characteristically calm: “We are pleased with our first quarter results, with strong bottom line performance, even with somewhat weaker macroeconomic conditions.” That sentence will get reprinted in every wire story today. It is not the sentence that matters.

The sentence that matters is buried further in, and it comes from the call rather than the release itself: Hilton has expanded its AI Planner — currently in beta on Hilton.com — and the underlying language model on the consumer-facing experience is Anthropic’s Claude. Layer on the broader partnerships announcement (Anthropic, Google, and OpenAI, all named on today’s call), and you have the cleanest articulation I have seen, anywhere in hospitality, of what an incumbent’s AI strategy actually looks like when the incumbent is serious.

I want to walk you through what today’s print says, what the AI Planner story actually means (as opposed to what the press release wants it to mean), and what franchisees and asset managers should be asking Hilton corporate about this week — before Marriott’s Q1 lands in early May and the comparison becomes unavoidable. This is an Operator piece, and the Operator read is blunt: Nassetta is using AI to defend distribution, not to disrupt it. That is a more important sentence than anything in the headline of the earnings release.

What today’s print actually said

Let’s get the numerics out of the way, because they frame everything else.

Hilton reported system-wide RevPAR of +2.5% on a currency-neutral basis for Q1 2025. That is a number that, in any other twelve-month period over the past decade, would have been received as a perfectly fine quarter — not a barnburner, not a disappointment. In the current macro environment, with “somewhat weaker” conditions (Nassetta’s framing, not mine), +2.5% is a signal that the brand is holding rate and occupancy in a market that is starting to wobble at the edges. Full-year guidance is flat to +2.0%. Read that twice. The midpoint of the range is +1%. The company is telling you the rest of the year is going to be harder than the first quarter was, and they want you to know that now.

Net unit growth came in at 20,100 rooms in the quarter. That is the operational metric a public hotelier always wants to put next to a softening RevPAR number, because pipeline growth is the thing that keeps fee revenue going up even when same-store comp slows. It is also, importantly, the metric that tells owner-operators and developers that the brand promise — distribution, loyalty, technology — is still pulling new construction and new conversions in. You can read the full release at the Hilton stories site (https://stories-editor.hilton.com/wp-content/uploads/2025/04/Q1-2025-Earnings-Release_APR2025.pdf); the rooms-growth tables are in the supplementals.

So the print itself is: solid quarter, softer guide, healthy pipeline. The market will trade on this for about ninety minutes. Then it will move on. What it should not move on from — what every Operator reading this should rewind to — is the AI section of the prepared remarks.

Why “AI Planner in beta on Hilton.com” is the key sentence

The AI Planner, as a product, is not new. Hilton first surfaced it publicly back in March in a press release that ran in the hotel trade press; Hotel Dive’s coverage from earlier this spring (https://www.hoteldive.com/news/hilton-launches-generative-ai-agent/814424/) walked through the initial framing — a conversational interface inside Hilton.com that helps prospective guests describe a trip in natural language and surfaces relevant properties, room types, and ancillary recommendations.

What is new today is two things, and they are easy to miss if you read the trade coverage the way it gets written, which is “Hilton expands AI Planner.” That is technically true. It is also a substantial undercount of what Nassetta said on the call.

First: he attributed the consumer-facing experience to Anthropic. That attribution was not in the March release. The March release was vendor-vague. Today’s call commentary names the LLM provider, on the record, on an earnings call. That is a deliberate disclosure. CEOs do not name vendors on earnings calls by accident. Hilton wants the buy-side, the trade press, and — more interestingly — its franchisees to know that the company has picked a model partner and is willing to stand behind that choice. Hotel Dive’s writeup of today’s vendor partnerships announcement (https://www.hoteldive.com/news/hilton-bets-ai-with-vendor-partnerships/818938/) goes through the Google and OpenAI pieces; Customer Experience Dive’s parallel coverage (https://www.customerexperiencedive.com/news/hilton-bets-ai-with-vendor-partnerships/818978/) frames it as the strategic posture I am about to lay out.

Second: he framed the AI Planner as living on Hilton.com. Read that part again. The product is not being released as an SDK, not being licensed to OTAs, not being embedded inside Booking.com’s interface, not being pushed out to Apple Intelligence or Google Assistant or any of the other agentic-booking surfaces the travel-tech press is currently writing breathless think pieces about. It lives on hilton.com. Full stop.

That is the strategic frame. AI is being deployed to make the brand’s owned channel smarter — so that when an agentic booking platform shows up at Hilton’s door and offers to “intermediate the customer relationship for a small commission,” Hilton can credibly answer: our channel is already conversational, our channel is already personalized, our channel is already as smart as your channel, and our channel is where the loyalty points live. Defend, don’t disrupt. The phrase will repeat in this column.

The three-vendor LLM mix, decoded

The vendor partnerships announcement is the part that the headline writers will get partly right and mostly wrong. So let me unpack it the way an Operator should think about it.

Hilton announced partnerships with three foundation-model providers: Google, OpenAI, and Anthropic. The trade-press default frame is “Hilton hedges its AI bets across the three biggest labs.” That framing is fine as far as it goes; it is also, in my view, the lazy read.

The non-lazy read: these three vendors are doing three different jobs.

Anthropic is powering the consumer-facing AI Planner on hilton.com. That is the brand-voice surface. It is the surface where, if the model says something weird or off-brand, you get a New York Post screenshot. It is the surface where tone, refusal behavior, hallucination tolerance, and safety guardrails matter more than raw benchmark performance. Hilton picked Claude for the consumer interface. That is a tone-and-safety choice, and anyone who has worked with the Claude family of models will recognize why.

Google’s role, based on the partnership framing, looks heavier on the data and search-and-retrieval side — the operational layer rather than the consumer surface. Think property data, trip planning context, image search, mapping. The kind of work where Google’s existing infrastructure is hard to replicate and where the integration with Workspace and search makes sense as an enterprise-level pipe.

OpenAI’s role is the most interesting, and the one I will be watching most closely. OpenAI is the player most aggressively positioning itself as the consumer-facing agentic-booking layer of the future — the ChatGPT-as-travel-agent narrative. By being inside that tent, Hilton gets early visibility into how OpenAI is going to expose hotel inventory in an agentic flow, and the optionality to plug Hilton.com data and loyalty into that surface on terms Hilton helps define. This is the partnership most clearly oriented at managing the channel-conflict problem before it becomes a channel-conflict crisis.

Put it together and the three-vendor mix is not “Hilton hedges.” The three-vendor mix is “Hilton is building a strategy where the consumer surface, the operational backend, and the agentic-booking optionality each have a different best-in-class partner, and the brand sits at the center of all three.” That is a more sophisticated posture than the trade press is giving it credit for, and it is the part that, in a later piece we publish on the QSR-side AI scaling story (/blog/posts/the-mcdonalds-ai-drive-thru-from-apprente-to-google-a-five-year-case-study), I want to come back to — because the QSR analog here is going to look very different, and the contrast is instructive.

Why Marriott’s AI Incubator looks weaker by comparison

I want to be careful here because Marriott has not yet printed Q1 — that lands in early May, and I will write the Marriott column the day it does. But the comparison is unavoidable today, and I would rather lay the frame down now than retrofit it later.

Marriott’s public AI posture, to date, has centered on what the company calls an “AI Incubator.” The Incubator framing — and I am giving this the most generous reading I can — is a portfolio approach. Marriott is exploring AI use cases across the business, partnering with vendors and startups, running pilots, and reporting progress as the pilots mature. The framing is fundamentally R&D-coded. It says: we are figuring this out.

That is not a bad posture in 2023. It is not even a bad posture in early 2024. In April 2025, with Hilton on an earnings call naming three foundation-model partners and pointing to a live consumer product on its owned channel, the Incubator framing reads as a brand still in scoping mode while its largest competitor is in deployment mode.

To put it more bluntly: Hilton today named the LLM behind the consumer surface, named the model providers behind the operational stack, and pointed to a product (the AI Planner) that real travelers can actually use right now on the company’s website. Marriott has talked about an Incubator. Those are not the same thing. One is a deployed product strategy. The other is a press release strategy.

I will be watching Marriott’s Q1 print closely to see whether the Incubator framing gets reframed as something more concrete. If Tony Capuano gets on his call and names model providers, names live products on the Marriott website, and gives a credible answer to the channel-defense question Hilton is now setting the bar on, then the gap closes. If the framing stays at the Incubator level, the gap widens, and asset managers should start asking pointed questions about why one of the two largest brand portfolios in lodging is moving slower than the other on the single most important technology shift of the decade.

The Operator point here is not about which brand wins the AI horse race. The Operator point is that AI strategy, for an incumbent, is a distribution-defense strategy. The brand that gets to the conversational, personalized, agentic-ready owned channel first is the brand that pays less in distribution costs over the next decade. Hilton just told the market that brand is going to be Hilton. Marriott has two weeks to respond.

The 25% rate-control framing — the strategic line every Operator should re-read

There was a line on today’s call that, more than anything else, told you how Nassetta is thinking about distribution. The framing went something like this — and I am paraphrasing the prepared-remarks language: Hilton, by virtue of its share of the lodging market and the brand-managed inventory that share represents, has meaningful influence over rate-setting across roughly a quarter of the market. The implication being: when Hilton makes a rate-discipline decision, a measurable share of the industry moves with it.

I am going to set aside the question of whether the 25% number is the right way to measure influence (it is a defensible number; it is also a number that other people will quibble with). The framing is what matters. Nassetta is publicly anchoring Hilton’s strategic identity as a rate-setter rather than a rate-taker. That is an extraordinary thing to say on an earnings call, and it is the kind of thing you only say if you are very confident in the brand’s pricing power going into a softer macro environment.

Connect that to the AI Planner. If you are a rate-setter and you have a softening macro, the last thing you want is for your distribution to get intermediated by an agentic-booking platform that compresses your pricing power by surfacing your inventory next to competitors’ inventory in a context the platform controls. The AI Planner on hilton.com is, in that frame, a moat. It is a way of ensuring that when a traveler with intent comes to plan a trip, they come into a Hilton-controlled context first, where Hilton’s pricing logic, Hilton’s loyalty offers, and Hilton’s upsell paths are the dominant signals — not a third party’s ranking algorithm.

The 25%-of-the-market framing and the AI Planner framing are the same strategy expressed two different ways. The strategy is: maintain pricing power; do not let the distribution stack erode that pricing power; and use AI to fortify the owned channel before agentic intermediaries get good enough to take rate compression as a tax. Read in that context, the AI announcement is not a tech announcement. It is a margin-defense announcement.

What today says about the rest of 2025

A few things follow if the frame above is correct.

First, expect Hilton to expand AI Planner functionality on hilton.com aggressively over the rest of the year. The product is in beta now. Beta in 2025 typically means “we are publicly admitting it is not finished, but we want the data from real users.” Expect feature releases — likely on a quarterly or near-quarterly cadence — that move the product from “trip-planning assistant” toward “end-to-end booking and upsell concierge.” Each release will narrow the functionality gap between hilton.com and what an agentic-booking platform could plausibly offer.

Second, expect the vendor disclosures to firm up. Today’s call attributed the consumer surface to Anthropic. Future calls will, I would bet, attribute specific functional layers to specific vendors as the architecture stabilizes. CEOs name vendors when they are confident in the choices; the confidence-building period is now.

Third, expect the loyalty program to get pulled deeper into the AI surface. Hilton Honors is the company’s actual moat. Any AI experience on the owned channel that does not, within a couple of release cycles, become deeply personalized to a logged-in Honors member is leaving money on the table. Watch for the integration of stay history, point balances, and tier-specific offers into the Planner experience.

Fourth — and this is the speculative one — expect Hilton to start exposing limited, controlled APIs into the OpenAI agentic surface as that ecosystem matures. The partnership announcement makes that pathway available. Doing it on Hilton’s terms is the only way to participate in agentic booking without being intermediated by it. The interesting question is what Hilton requires in exchange — loyalty enrollment? direct-billing relationships? brand-managed display in the agentic UI? — and which of those terms OpenAI is willing to accept.

What franchisees should ask Hilton corporate this week

If you are a franchisee or asset manager with Hilton-flagged properties in your portfolio, here is the short list of questions I would put on the next franchisee advisory call. These are the questions I would be asking if my fee structure and my distribution costs were going to be shaped by the strategy Nassetta laid out today.

How does AI Planner rank properties? When a traveler types in a destination and dates, what is the underlying ranking logic that surfaces specific properties? Is it loyalty-tier-weighted? Performance-weighted? Brand-segment-weighted? Owners have a right to know whether their property is being shown to the most relevant guests, and the answer to that question is increasingly going to be a function of LLM ranking logic rather than traditional sort criteria.

What is the property-data input requirement? AI Planner can only be as good as the property descriptions, photos, amenity tagging, and event-and-meeting capacity data that flow into it. If corporate is going to lean on AI Planner as a conversion surface, owners need to know what the data hygiene requirements look like and what investment is required at the property level to keep the inputs current.

How does AI Planner interact with direct-booking promotions? If a franchisee runs a rate promotion, does AI Planner surface it? With what weighting? On what timeline? The owned channel is the channel where promotion economics are best for the owner. If the AI surface does not foreground promotions effectively, that economic advantage erodes.

What is the personally identifiable information posture? AI surfaces process guest queries that can include trip purpose, traveling-with-children context, accessibility needs, and other sensitive information. Where does that data live? Who has access to it? How is it protected? Hilton corporate will have an answer; owners should make sure they have heard it.

What happens to AI Planner in the agentic-booking world? If, in eighteen months, a traveler is planning trips through an agentic interface that connects to Hilton.com through an API, does the AI Planner surface persist on hilton.com, or does the brand experience get exported to the agentic surface? The answer to that question shapes whether the brand-experience moat is durable or fragile.

These are the questions that, in our later coverage of the hotel-tech stack (/blog/posts/desk-review-sevenrooms-vs-tablecheck-two-reservations-platforms-built-on-opposite-bets), I would expect the conversation to move toward as the AI-distribution story matures. Today’s print is the opening move. The rest of 2025 is going to be a series of clarifications.

Operator takeaways

• AI Planner on Hilton.com is a distribution-defense strategy, not a tech demo — read it as a move to anchor the customer relationship inside the brand’s owned channel before agentic-booking platforms can intermediate it.
• The three-vendor partnership (Anthropic for the consumer surface, Google for the operational backend, OpenAI for agentic-booking optionality) is more sophisticated than the trade press is treating it; the brand sits at the center of all three.
• Nassetta’s 25%-of-the-market rate-control framing and the AI Planner roll-out are the same strategy: defend pricing power going into a softer macro by fortifying the owned channel.
• Marriott’s “AI Incubator” framing now looks R&D-coded compared to Hilton’s deployment-mode posture; the May Q1 print is Marriott’s chance to close that gap, and asset managers should watch for whether Capuano names model providers and live products.
• Franchisees should be asking corporate about AI Planner ranking logic, property-data input requirements, promotion-surfacing behavior, PII handling, and agentic-API exposure — these are the questions that determine whether the brand’s AI moat is also the owner’s moat.

What to watch on Marriott’s print

Marriott reports in early May. I will write the column the day they print. Three things will tell us where the AI gap is heading.

One: does Capuano name his model providers? If yes, the gap closes meaningfully. If no, the gap widens.

Two: does Marriott point to a live, named, consumer-facing AI product on the marriott.com surface? An Incubator portfolio is not a product. A live product on the owned channel is a product. The presence or absence of one in the prepared remarks will tell you everything you need to know about where Marriott is in the deployment curve.

Three: does Marriott address agentic-booking distribution explicitly? Hilton telegraphed today that it is thinking about agentic booking as a structural shift in distribution, and that it is taking concrete steps to position the brand for it. If Marriott’s prepared remarks treat agentic booking as a curiosity rather than a structural force, that is a posture problem.

The honest read is that Hilton has set the bar today. Marriott has two weeks to clear it. The interesting question for everyone else in the industry — the IHGs, the Hyatts, the Choice Hotels, the Wyndhams — is what the bar looks like by their next print, because the standard that Hilton just set is the standard the buy-side is now going to hold every brand to.

That is the Operator read on today. Hilton did not just have a fine quarter. Hilton announced that it intends to own the AI agent at the consumer surface, on its own terms, on its own channel, with a defined model partner and a defined strategic frame. That is the difference between deploying AI and owning the AI agent. The difference is going to be worth a great many basis points of distribution cost over the next decade, and the brand that gets to “owning” first is the brand that pays less in the tax on its own demand.

I closed the laptop, put it back in the bag, and went downstairs for coffee in the lobby — paid for, naturally, with Hilton Honors points. The future of how I made that purchase, and how I will make the next one, is the thing Nassetta is trying to lock in this year. He told us as much today. The Operator question is whether the rest of the industry was listening.

— Naomi covers hotel F&B and operations. Tips: tips@tabletransfers.com.