Tracking your hotel in AI answers

AI answers are not a fixed ranking. Assistants update how they read the web, competitors change their websites, and the same question can return different hotels on different days.

That means one good result can be luck, and one bad result can be noise. What you can trust is the direction over time: whether your hotel appears more often this quarter than last, and whether it moved after a specific change.

Keep a simple spreadsheet. For every question you ask, on every assistant, record five things.

1. Mentioned or not. Did the answer name your hotel? This is your mention rate when you add it up across all answers.
2. Position. If you were named, were you first, second, or further down the list?
3. Who was named instead. Write down every other hotel in the answer. Over time this becomes your real competitor list in AI answers.
4. What was cited. If the assistant showed sources, note whether it pointed to your own website, an OTA listing, a review site or a press article.
5. How you were described. Copy the sentence the assistant wrote about you. Changes in this sentence often show which part of your website it is reading.

A simple layout works well:

Tracking only works if each round is comparable. Choose a fixed set of questions that reflect your real guests, for example a quiet weekend, a first visit to your city, a family stay and a stay near the landmark or station your guests use, and ask exactly the same wording every time.

Add a new question only as a separate line, and do not replace an old one. Otherwise you cannot tell whether a change came from your hotel or from the question.

Monthly is enough for most independent hotels. AI answers do not change meaningfully day to day for a single property, and checking more often mostly adds noise and work.

Run an extra round a few weeks after any significant change to your website, such as unblocking AI crawlers, adding Hotel schema markup or rewriting your main pages. That gives you a before and after for the change itself.

Hand checks are easily distorted. A few habits make them more reliable.

• Start a fresh conversation for every question, so earlier answers do not influence later ones.
• Use a logged-out or private session where you can, because some assistants personalize answers based on your history.
• Ask each question more than once and count how many times you appear, rather than recording a single yes or no.
• Check more than one assistant. ChatGPT, Gemini, Claude and Perplexity read the web differently, and a hotel can be visible on one and absent from another.
• Look at the trend, not one round. A single month that moves up or down is not a result on its own.

Your own mention rate means more when you see it next to the hotels that compete for the same guests. From your spreadsheet, count how often each competitor was named across all answers in a round.

If one hotel appears far more often than you do, look at its website. Check whether it states clearly who it is for, whether it publishes structured data about its rooms and location, and which sources the assistants cite when they recommend it. This usually points to the specific gap on your own site.

Change one thing at a time where you can, note the date, and compare the rounds before and after. Look at three signals together:

• whether your mention rate rose across the same questions,
• whether your position in the lists improved,
• whether more citations point to your own website rather than to an OTA.

The third signal matters commercially. A recommendation that cites an OTA describes you through someone else's page, so moving citations to your own domain is a real gain even before your mention rate changes.

A monthly hand check is a good discipline and costs nothing. Its limit is scale. A reliable picture needs dozens of questions, on all four assistants, each asked several times, and the effort grows quickly.

That is the part a structured measurement handles. Our sample report shows what a complete one looks like, using a demo property: the questions asked, how often the hotel appears on each assistant, who is recommended instead, which sources are cited, and a prioritized list of fixes.