AI visibility can show where a hotel appears, how it is described, which sources shape the answer, and whether the traveler is sent to the hotel or an intermediary. It cannot show which real guests asked about the property and which ones booked. I think that second question will become much more valuable than the first.
I have spent the last stretch measuring how hotels show up when travelers ask AI assistants where to stay. The work has made one thing very clear: measuring the public answer is useful. It is also incomplete in exactly the same way most hotel commercial data is incomplete, and we already hold the evidence that completes it.
You can see what happened. You cannot always see why.
What can AI visibility measurement tell a hotel today?
A good visibility analysis can answer several important questions.
Does the hotel appear when a traveler asks about the destination? Does it appear only when the prompt already names its competitive set? Is the property described accurately? Which public sources keep shaping the answer? Does the traveler get routed toward the hotel, an intermediary, a review page, or nowhere useful at all?
Those questions matter because AI assistants are becoming another layer of discovery. A traveler can describe the trip they want and receive a short list before they visit a hotel website, a search result, or a booking marketplace.
The information available at that moment can shape the consideration set.
That is worth measuring. It should be measured rigorously, across repeated prompts and against more than one denominator. A single screenshot is noise. A single score without a declared comparison set is incomplete.
But even when the measurement is done correctly, it observes the hotel from the outside.
Why is source tracking useful but incomplete?
Knowing which sources shape an AI answer is useful.
If a property's own website is repeatedly cited, the hotel has evidence that its content is helping define the public description. If third-party sources dominate, the commercial team can see who is carrying the narrative. The SIGIR 2026 study found substantial differences between the sources retrieved by the search surfaces it tested. Its review of prior research also notes the importance of third-party reviews for product and service questions. For a hotel, that makes source tracking actionable: it can show whether the public answer is being shaped by the property, review ecosystems, or other outside sources. If the cited information is wrong or stale, the team knows where to investigate.
But source tracking is not the final insight.
It tells you what information the assistant found. It does not tell you what the guest wanted badly enough to call, email, open a chat, or request a proposal.
Those are different evidence sets.
The public model sees the web. The hotel sees the guest.
Today, those two views are usually analyzed separately, if the second one is analyzed at all.
What can public models never observe?
A public model cannot see the full demand conversation happening inside a hotel.
It cannot see the guest who asked whether the pool would be open during construction and then never booked. It cannot see the family who needed connecting rooms, received an uncertain answer, and disappeared. It cannot see the group planner who asked for a specific setup three times before choosing another property.
The hotel had those signals. They arrived through channels the hotel owns.
At most properties, they vanished. At the properties we work with, they did not.
This has been the central problem behind everything we have written at Anana. A PMS knows the reservation that exists. An RMS knows the booked demand it can price and forecast. Neither one was designed to preserve every question, hesitation, objection, and unmet need that appeared before the booking.
AI visibility adds another version of the same gap.
The outside world can now be measured with increasing detail. We can see where a property appears, which answer it enters, and which source gets cited. But the commercial meaning remains limited until that outside-in view is compared with the demand signals happening inside the property.
Why does “who asked versus who booked” matter for AI visibility?
Imagine a resort is rarely recommended for family trips in its region.
That is an outside-in finding. It tells the team that the public information layer does not strongly associate the property with family demand.
Now imagine the resort's calls and chats contain repeated questions about connecting rooms, children's programming, meal plans, transfers, and whether the beach is safe for young swimmers.
That is an inside-out finding. It tells the team the demand exists, even if the public model does not recognize the property for it.
The gap between those two views is commercially useful.
Maybe the website does not answer the questions guests ask most. Maybe third-party descriptions are incomplete. Maybe the hotel is attracting family demand but losing it in the conversation. Maybe the property should not pursue that segment at all.
AI visibility alone cannot tell you which explanation is true.
The hotel's own interaction data can supply the evidence needed to investigate.
Where do I think this category goes next?
I do not think the category ends with a visibility score.
Measurement is the entry point. Hotels need to know whether they appear, how consistently they appear, what is said about them, and where the traveler is routed. That work is real and valuable.
It is also the easy half. The hard half is comparison.
Compare the public sources shaping the hotel's AI perception with the questions real guests ask. Compare the needs associated with the property online with the needs expressed on the phone. Compare regional discovery gaps with the demand that reaches the reservations team anyway. Compare the travelers an assistant might send toward the hotel with the travelers the hotel actually converts.
That is where AI visibility becomes commercial intelligence.
Both halves are real. We capture and structure guest interactions across voice, email, and chat today, at scale, in production. We measure how properties appear across public models today. What no one has finished, us included, is the fully automated version that reconciles every public model against every owned interaction without a human in the loop.
I would rather say that plainly than pretend a measurement dashboard has already solved the harder problem.
Why should the commercial team own this question?
AI visibility will often enter the hotel through marketing because it looks like a search problem.
Marketing should care. But the business outcome crosses revenue, sales, reservations, and operations.
If the hotel is recommended for the wrong traveler, that affects conversion. If the answer routes demand through a costly path, that affects channel economics. If the public description misses a service guests repeatedly ask about, that affects positioning. If a visibility gap is actually caused by an operational uncertainty, better content alone will not fix it.
This is why I see AI visibility as a commercial question.
The public answer is one signal. Booking data is another. Guest interactions contain the explanation between them.
No single team owns that full chain today.
Can hotels reliably optimize AI visibility today?
Not yet in the way the word “optimize” usually implies.
The SIGIR 2026 research found that the generative-search systems it tested were less consistent across repeated runs than traditional search and less robust to cosmetic query edits. The paper also notes that the effectiveness of generative engine optimization remains contested.
That does not mean hotels should do nothing. It means this is still a field of running experiments, not a channel with a durable lever that moves a stable ranking on demand.
The responsible posture today is to make a change, measure it across repeated prompts, and ask whether the result exceeds the normal noise of the system. No one should promise a magic button for moving an AI ranking. The durable lever has not arrived yet.
That is another reason not to stop at outside-in measurement. Hotels can experiment with public visibility while preserving the owned demand evidence that public models will never see.
What should hotels do now?
Start with measurement, but do not stop there.
- Measure repeatedly. One answer is an example, not a benchmark.
- Declare the denominator. Compare the hotel against its direct set, its region, and the position it wants to own.
- Inspect representation and routing. A mention can still be inaccurate or send the traveler somewhere the hotel does not control.
- Preserve owned demand signals. Calls, emails, chats, and inquiries contain the reasons behind conversion and loss.
- Look for the gap. Compare what the public information layer says about the hotel with what real guests keep asking.
The hotels that do this well will not treat AI visibility as a new vanity metric. They will treat it as another evidence set inside the commercial investigation.
The outside view is not the whole truth
I believe AI assistants will influence more hotel discovery and more booking paths. Some will send travelers to hotel websites. Some will route them through intermediaries. Some will answer the question so completely that the traveler delays the click altogether.
Hotels should understand that environment.
But knowing that your property appeared in an answer is not the same as understanding demand.
The more important question is still the one hotel systems have struggled to preserve for years: who asked, what did they want, and why did they book or walk?
AI visibility shows perception. The next layer of hotel intelligence connects that perception to the conversations and outcomes the property already owns.
That is the category we are building at Anana. If you are working through the same question, I would love to compare notes.