How to get cited by ChatGPT as a hotel
An assistant can send you visitors without ever writing your name, and it can write your name without sending anybody. These are two separate outcomes with two separate fixes, and the only way to tell which one you are missing is to count them apart.
AI sessions measured across the sites we track, last 30 days.
Quick answer
Across the corrected measurement window, 29 of the 57 questions we track belong to one cenote park in Punta Cana, run on 5 AI surfaces, which is 145 of the 285 measurements. The site showed up in 74 of them. It was linked as a source 73 times and named inside the answer text only 23 times, and 51 of those 74 appearances were a bare link with its name absent from the answer. Getting cited by ChatGPT is therefore two jobs and not one: earn the link, then earn the mention.
Being named and being linked are different wins
Most hotels ask a single question, which is whether the assistant knows them. The measurement splits that into two answers with very different value.
A mention is your name inside the sentence a guest reads. It survives the zero click case, where nobody visits anything and the reader still walks away knowing the place exists. A link is your URL in the source list under the answer. It brings a real session, and because ChatGPT tags its outbound links with utm_source=chatgpt.com, that session lands in your analytics as something you can count.
In the window above the two counts were almost opposites. The site was linked as a source on all 5 surfaces we run, and named in the answer text on only 2 of them. Google AI Mode named it 14 times and ChatGPT 9 times. Claude, Gemini and Perplexity named it zero times, and all three still linked to it: Perplexity 20 times, Claude 16 and Gemini 15.
Read as one number, that site looks well covered. Read as two, it has a link problem it does not have and a naming problem it very much does.
What each of the two numbers responds to
The link responds to your own pages. An assistant links a URL when the page answers the question it was asked, in words it can lift, near the top, with the practical facts written as text rather than baked into an image or a booking widget. That work happens on your site and you control all of it.
The mention responds mostly to other people's pages. When the model writes a name in prose it is repeating a name it has seen described elsewhere, in guides, roundups, review sites and destination pages. In the same window, the answers about that site cited 784 sources across 271 different domains. Its own pages supplied 73 of them, so more than 90 per cent of what the models read to build those answers was written by somebody else.
That is why a hotel can fix its website, watch the link count rise, and see the mention count stay flat. They are fed by different things. We cover the on-site half in content that AI cites and the off-site half in off-site mentions.
Measure it before you change anything
Read all of the above as mechanics and not as a forecast. The 7 domains in this pass are our own travel websites rather than hotel clients, so the split between naming and linking is the transferable part, and the hit rate is not. It is also one pass, and a single reading proves nothing: what makes a number an argument is repeating it.
A pass in reporte.ai is a fixed list of questions, run on every surface, stored with the full response text so you can reread why you did or did not appear. The window we describe here holds 285 measurements across 7 domains with zero errors, which matters because a failed run that silently returns nothing looks exactly like an answer that did not mention you. Ours did exactly that once, which is why the window is defined as the last clean run of every cell rather than as a calendar day.
Alongside it, the pixel on your site counts the sessions each assistant actually sends. That is the pair you want: what the models say about you, and what that turns into. See prompt tracking and AI traffic for how each half works.
FAQ
Two different things that our tracking separates. One is your name written inside the answer text. The other is your URL listed as a source the assistant used. In the corrected window the site we measured was linked 73 times and named 23 times across 145 of the 285 measurements, so the two counts are not interchangeable.
The mention travels further, because a reader who never clicks still leaves with your name. The link is what brings the session and what you can measure, since ChatGPT tags its outbound links with utm_source=chatgpt.com. A site that gets one without the other is only halfway there, and 51 of the 74 appearances we measured were a link with no mention.
The site described here accounts for 29 of the 57 questions in the pass, run against 5 surfaces, which is 145 of the 285 measurements. Fewer questions than that and a single lucky answer moves your whole score.
The 5 we run are ChatGPT, Claude, Gemini, Google AI Mode and Perplexity. They behave differently enough that checking one tells you almost nothing about the rest: in this window the site was named on 2 of the 5 and linked on all 5.
Yes, when the citation is a link. Across the sites we measure, ChatGPT sends around nine of every ten AI sessions and more than twenty times the next assistant, and the panel at the top of this page carries the live 30 day counts. Our pixel files each visit under its own assistant instead of dumping it into direct traffic.
No, and that matters more than the numbers. The 7 domains in this pass are our own travel websites, not hotel clients, so what travels is how the measurement behaves and not how a property performs. It is also a single pass, and a single reading proves nothing: your own baseline is the only figure worth acting on.
We do not publish a promised timeline, because we have not measured one we would stand behind. What we can give you is the baseline today and the same pass repeated after your changes, so the movement is measured rather than assumed.