What share of voice means for a hotel
Detection rate answers one question: out of the traveler questions you track, how many name your hotel. Share of voice answers a harder one: on those same questions, who got named instead of you, and how often.
The arithmetic is worth being explicit about because it surprises people. One answer can name three hotels, and each of them gets a mention. The shares are therefore independent of each other rather than slices of a fixed pie, and they can total well above 100% across a set of competitors. If you see share of voice drawn as a stacked bar adding to exactly 100%, the chart is misrepresenting the measurement.
Why detection rate alone hides the real problem
Picture two hotels that both score 10% detection. In the first case, 90% of the answers name no hotel in that category at all: the engine is hedging, or the question is too vague for it to commit. In the second, 90% of the answers name the same competitor. Those are the same number and completely different situations.
The first is a market gap and the work is to be the first clear answer to a question nobody currently owns. The second is a competitive loss and the work is to appear in the places that competitor already appears. Without a comparison you cannot tell which one you are in, so you cannot tell which of those two jobs to do.
Co-citation: which hotels AI treats as your substitutes
When your hotel is named, pay attention to who is named in the same breath. That set is the engine's working answer to "what else would this guest consider", assembled from everything it has read rather than from your own positioning.
It often disagrees with the comp set a revenue manager would write down. A hotel that prices against three nearby properties of the same class may find the answers pair it with somewhere cheaper two neighbourhoods away, because the engine is grouping by the trip the guest described rather than by rate or star rating. That disagreement is useful information about how guests arriving through AI are framing their choice.
How to pick the right competitors to track
Derive them from your own answers. Run the questions, write down every hotel named in place of yours, and count. The ones that recur are your real competition for the citation, whatever your rate shop says.
There is a practical limit worth planning around. Tracked competitors cost measurement capacity, so plans cap them: one on Radar, three on Essential, five on Pro and Group. That cap is a feature in the sense that it forces a decision. Three competitors you chose from evidence will teach you more than ten chosen from habit.
Doing this by hand, and where it breaks
A general manager can do a usable version in twenty minutes: one question, two engines, note which hotels appear. That is enough to find out whether you are in a market gap or losing to a specific rival, which is the main thing.
Where it breaks is repetition. These answers are not deterministic. Ask the same question twice and you can get two different sets of hotels, so a single run is closer to a coin flip than a measurement. Getting a number you would act on means asking the same questions repeatedly over time, and that is the point at which the manual version stops being practical rather than the point at which it stops being accurate.
What a real market looks like
Our Prague destination report is a worked example of share of voice at city scale. Across 320 real answers to 46 non-branded questions about where to stay in Prague, Four Seasons Hotel Prague was named 91 times. The rest of the top ten ran from 60 mentions down to 29. No single hotel came close to owning the market, and the gap between third and tenth was narrower than most hoteliers would guess.
Two hotels in that top ten are Staylight customers with published case studies, Mosaic House at 51 mentions and Golden Well at 49, and we disclose that on the report itself rather than presenting the list as if we had no relationship to it. The snapshot is observational: it shows who the engines named in a particular week, and it does not demonstrate that being a customer caused the placement.
What to do when a competitor owns a question
Find out why they are there before deciding what to do. Read the citations in the answers that name them. Usually the competitor is not winning on their website: they are winning because two or three editorial or local sources that the engine trusts mention them in this context and do not mention you.
That makes the work specific rather than general. You are not trying to be better than them overall. You are trying to be present in the handful of places the engine consults for that particular question. Our guide on why AI recommends OTAs instead of your hotel covers how to pull that list of places out of your own answers.