By Michael Goldrich, Vivander Advisors
Cloudbeds ran 810 prompts about 145 hotels across six destinations through ChatGPT, Perplexity and Gemini, then traced where the answers came from. Online travel agencies supplied 55.3 percent of the citations. Hotel websites supplied 13.6 percent. Travel blogs outranked hotel websites as a source. So did community forums once you count Reddit and TripAdvisor together.
That ratio inverts 15 years of digital strategy. Your website has been the record of truth about your property. To an AI answering a traveler’s question, it is a minority source.
Hotels have spent a decade working to reduce OTA dependence. That work still makes commercial sense. Something else happened while the industry focused on shifting share. Those listings quietly became the reference material machines read when deciding which properties to put in front of a traveler.
The distinction worth holding onto: the citation and the booking are two separate transactions. An accurate Expedia profile can feed the recommendation that sends a traveler to your own site to book direct. Improving that profile costs you nothing on the commercial side. It improves the raw material the model is working with.
Consider what happens when your sources disagree. Your website says check-in is 3 p.m. Booking.com says 4 p.m. Your Google Business Profile never had the field filled in. The model resolves the conflict by trusting whichever source it encounters most often, or by dropping the detail entirely. Properties with clean, matching information across every listing give the AI something it can state with confidence.
Here is the work, and none of it requires budget approval.
- Build your fact sheet. One page, roughly ten facts, all numbers. Check-in and check-out times. Pet policy with the weight limit and the fee. Parking with the nightly rate and whether in-and-out is included. Square footage by room type. Restaurant and pool hours. Meeting space by the square foot. Distance to the airport and the convention center. Accessibility features. Any resort fee. This page becomes the reference for everything downstream.
- Compare five sources against it. Your website, your Google Business Profile, Booking.com, Expedia and TripAdvisor. Facts down the left, sources across the top, then fill in the grid. Most properties find somewhere between six and fifteen contradictions on the first pass. Correct them at the source rather than in a spreadsheet.
- Replace adjectives with specifics. “Spacious guestrooms” gives a model nothing to work with. “425-square-foot guestrooms with blackout curtains and double-pane windows on the courtyard side” gives it a reason to recommend you when someone asks about quiet rooms downtown. Start with the three pages carrying your highest-value business.
- Test, then keep testing. Write ten questions a traveler would actually ask about your market. Run them through ChatGPT, Gemini, Perplexity and Claude. Record which platforms name your property, which skip it, and whether the details come back accurate. Save the answers. Run the same ten questions again next month so you are watching movement instead of a single snapshot.
One more surface deserves attention. Your review responses sit on pages these models already read. “Thank you for your feedback” contributes nothing. A response that names your rooftop hours and describes what changed on the spring menu adds specific attribute data to a source the AI already trusts. Every response is a small deposit into the citation network.
The Cloudbeds study carried a finding worth sitting with. Brand-affiliated properties appeared in 72.4 percent of recommendations. Scale still helps. Independent operators are not locked out of this, because the levers that matter most sit at the property level: your Google Business Profile, your listing accuracy, your review responses, your local press coverage. A corporate digital team cannot maintain those from a distance. No franchise agreement covers them.
Ask your team who owns this today. The answer is usually some version of “we all sort of do,” which means the check happens when someone remembers. Schema breaks during a website update and nobody validates it. A pet policy changes on the website and stays wrong on three OTAs for a year. Ownership has to belong to one person with protected time on the calendar.
The first pass through this takes an afternoon. Keeping it accurate is the ongoing part, and it belongs in the same category as your inventory and your rate parity. Facts about your hotel are now an asset that decays without maintenance.
Travelers stopped scrolling through ten links. They ask one question and act on one answer. What that answer says about your property is assembled from sources you influence today, most of which you have been treating as somebody else’s problem.
Michael Goldrich is founder of Vivander Advisors, a hospitality AI advisory firm focused on AI visibility, and the author of Invisible.
