Measuring AI mentions for hotels: reading the panel through the year

Why a hotel's month needs two comparisons

The panel stays fixed. The year it is asked in does not. A convention fills the town for a week, a festival brings a different kind of traveler, the schools break up, a renovation closes the pool. What an answer says about your property in such a month can move for reasons that have nothing to do with your site, and a reading taken without knowing that can be misread.

Month on month, a hotel's reading mixes two things: what changed about you, and what changed about the month. A fall between August and September may be your site, or it may be September. The comparison that separates them is the same month a year earlier, when the calendar was the same and your pages were not. Read each month against both, and write down which one you are talking about when you report a change.

Neither comparison means anything unless the questions behind it are the same. The panel itself, twelve lines by trip type, condition and place, is laid out in AI visibility for hotels, and how to measure AI visibility covers telling a real change from the noise between runs. This page is about the year: what to log beside each reading, and how to read a change that lands in the same month as a season.

The reading log: one entry a month

Keep one entry per month, written in the month it describes. The first lines come from the reading itself, scored the way our method scores any client: the same twelve questions, put once per reading to one assistant, Claude, and a visibility score from 0 to 100 that is the share of those questions in which the hotel was named. The rest is what the reading cannot know about your month.

month:             [YYYY-MM]
score:             [0 to 100]; last month [n]; same month last year [n, or no reading yet]
named / recommended / first:  [n] / [n] / [n] of [n] answers
own page linked:   [n] of the answers that named you
in town:           [the convention, festival, game or school holiday, with its dates]
at the property:   [renovation, a closed facility, a rate change, a policy change, with dates]
on the site:       [pages changed, with dates]
on the listings:   [changes to platform and map listings you made or noticed, with dates]
bookings:          direct [n]; guests who named an assistant [n]; AI referral visits [n]
search console:    AI Overviews and AI Mode impressions [n]; top pages [page: n]

The "own page linked" line is the link column from the panel's record, counted for the month. The bookings line comes from your booking engine and your analytics, as direct bookings from AI answers sets up. Write every date down. A note that says "renovation in the spring" cannot be matched to a reading; one that says the pool was closed from the third to the twenty-fourth can.

What to write beside each reading

Four kinds of note are worth keeping. Each one answers the same question a year later: was there a reason, outside the site, for this month to look the way it did?

The property notes deserve the most care, because they tell you when an answer was right to change. If your rate rose above the budget in a line such as "hotel in [city] under [amount] a night", an answer that stops naming you for that line is stating a fact. If the dog-friendly rooms were being refurbished, a fall on the dog lines that month is not a loss to chase. Mark those months, so that nobody spends next spring fixing a page that was never broken.

Reading a change that lands in the same month as a season

When a reading moves in a month with a lot going on, work through it in this order before deciding what it means.

1. Read the property notes first. If a fact changed that month (a closed pool, a new rate, a sold-out week), check whether the answers changed with it. A change that follows a fact is the answer being right.
2. Compare with the same month a year earlier. If the same dip or rise happened then, with nothing on your site changed in either year, the calendar is the likelier explanation.
3. Look at which lines moved. If only the lines about one venue or one kind of trip moved, look at what happened to that venue or trip that month. If every line moved on every assistant, look at access first: a new robots.txt rule or a challenge page can take a site out of all of them at once, and the crawler access check shows what your robots.txt allows.
4. Look at who else was named, the raw material of your share of voice. A property that appears in the others column only in convention weeks is a fact about those weeks, not about your site.
5. Hold the conclusion for a month. One reading is one dated answer per question. If the change is real it will still be there next month, and the notes will tell you whether the month was ordinary.

In the first year there is no same month to compare with, so say so in every report: "no reading yet for [month] last year". The first twelve months are a baseline in two senses: a first reading of each question, and a first reading of each month.

Where the hotel's own numbers sit beside the reading

Three numbers belong in the same entry as the reading, for the same month. Direct bookings by source come from your booking engine: the count of guests who answered "an AI assistant" to the "how did you find us" question, beside search, the platforms, the phone and returning guests. AI referral visits come from your analytics, with the assistants' hosts grouped into one channel as spotting AI traffic in Google Analytics describes. And for Google's own AI features there is Search Console.

Since August 31, 2026, Search Console's generative AI performance report has shown impressions of links to your site in AI Overviews and AI Mode, by page, country, date and device. It counts impressions only, and it has no query dimension, so it can tell you that your family room page was shown in more AI Overviews in one month than in the month before, and nothing about which questions brought it up. Its dates are in Pacific Time, which matters at the edges of a month if you log from another time zone.

Now the part to say plainly: none of this proves that an answer caused a booking. A season can move bookings and answers at the same time. A convention month can raise your mentions and your direct bookings together with nothing connecting the two, and a traveler who read your name in an answer and then phoned the desk shows up in no count unless someone asks. Put the numbers side by side, report them side by side, and do not draw a line from one to the other that the data cannot carry.

What the log gives you, after a year, is a fair comparison: each month against its own month, with the reasons written down. That is enough to see whether the work on your pages moved the answers, which is the question the panel exists to answer. Whether it moved the bookings is a question the log can inform and cannot settle. The hotel's own facts are in the overview for hotels.