The First Time an AI Told a Hotel Guest Where to Stay

I spent Thursday morning with a printout of Cloudbeds’ new study spread across the breakfast bar at a friend’s boutique in Lisbon, asking her revenue manager to guess which hotel ChatGPT had named when I typed “two-night stay near Time Out Market, quiet, under €220.” She guessed wrong three times. The model named a property she’d never heard of, whose Google rank she’d never thought to check. The report Cloudbeds released this morning — “The Signals Behind Hotel AI Recommendations” — is being read as a curiosity. It is not a curiosity. It is the opening shot in a battle over generative-engine optimization that will reshape hotel marketing budgets before the year is out.

The Cloudbeds line that everyone is quoting is Adam Harris’s: “Travelers no longer scroll through search results; they ask AI.” Slogans like that usually get filed under hype. This one is doing real work, because for the first time we have a vendor-funded study that systematically interrogates ChatGPT, Perplexity, and Gemini about hotels and reports back which properties they name, in what order, and why. (Confidence on the publication date here is medium — the trade press surfaced the report on July 1 and Cloudbeds appears to have staggered the public release; I am treating today as the operator-facing publication date.)

Why this study matters more than its press kit suggests

Cloudbeds is, of course, a property management system vendor with an interest in selling the next layer of marketing tooling to its installed base. Read the report charitably and the methodology still does something the industry has not done before: it treats generative engines as a measurable distribution channel and asks what signals correlate with being named.

The findings, as best I can reconstruct them from the ehotelier coverage, are the ones operators were quietly afraid of. Review velocity matters. Review recency matters more. Third-party mention density on travel publications and Reddit threads correlates with being recommended in ways that vanilla Google ranking does not. The presence of a structured, machine-readable property description — the kind of schema that PMS vendors have spent years cajoling clients to maintain — appears to matter materially. And brand affiliation, somewhat startlingly, matters less than I would have predicted: a well-described independent can beat a third-tier chain property in conversational queries.

Mark this as interpretation rather than reported fact, because the underlying dataset is Cloudbeds’ and the methodology is not yet third-party replicated. But if even half of it holds, the implications for where hotel marketing spend goes next year are not subtle.

The GEO budget shift is going to be ugly

Generative-engine optimization — GEO, the term has stuck — is going to do to hotel marketing budgets what SEO did to them in 2007, except faster and with fewer agencies prepared. The Cloudbeds report is the first piece of vendor-published evidence operators can point at when their CMO asks why a chunk of the Google Ads line should move toward structured-data hygiene, review-program acceleration, and earned coverage on the publications that LLMs are quietly training and grounding on.

What I expect to watch happen over the next two quarters: the metasearch line gets defended hard, because nobody can prove it’s dying yet. The display line gets cut, because nobody can prove it ever worked. And a new line item — call it “AI visibility” or “GEO” or whatever the agency pitches it as — appears in Q4 plans at independents first, then mid-scale chains, then majors. The majors will move last because their legal teams will spend a quarter arguing about whether feeding structured data to OpenAI constitutes a licensing relationship. The independents will move because they have nothing to lose and their breakfast bar conversations sound like mine in Lisbon did.

The contrast with the rest of the chain world is sharp. Skift’s June 4 read on Hilton’s approach — “Hilton’s AI strategy: less hype, more guest experience” — laid out a deliberately inward posture: AI for housekeeping routing, for staff scheduling, for guest-message triage. Nothing in that strategy points outward at how Hilton properties surface in a Perplexity query. The Cloudbeds study is going to land on a lot of brand-marketing desks this month that have, until now, been allowed to treat generative search as someone else’s problem.

What I’d ask Cloudbeds to publish next

A study that names which engines named which hotels would be more useful than one that aggregates across engines, because ChatGPT, Perplexity, and Gemini are not interchangeable surfaces; they ground on different corpora, rank with different signals, and answer with different prose. A follow-on study that breaks results out by engine — and that re-runs the same prompts ninety days later to test stability — would tell operators whether GEO is a discipline or a dice roll.

I’d also want geographic decomposition. The Lisbon test I ran with my friend’s revenue manager produced one recommendation set; the same prompt tuned for Austin produced an entirely different shape of answer, with chain properties surfacing more readily and independent inventory mostly disappearing. If generative engines weight differently by market, the GEO playbook is regional, not global.

The Cloudbeds Market Pulse dashboard gives operators a forward-occupancy read week to week. A GEO companion dashboard — share of LLM mentions for your property versus your competitive set — is the product I expect to see by Q1 2026. Whoever ships it first owns the category.

Two things to read alongside this when our editing schedule catches up: a forthcoming desk review of Toast’s restaurant-AI playbook, which is the parallel story on the F&B side, and an upcoming May piece comparing SevenRooms and Tablecheck’s data postures, because the data layer underneath the recommendation is going to matter more than the engine on top.

Cloudbeds did not publish a manifesto today. It published a starter pistol. The hotels that hear it as such will spend the back half of 2025 instrumenting for a channel that did not exist eighteen months ago. The ones that do not will discover, around Labor Day, that the channel was already deciding where their guests slept.

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