Your hotel can make the answer while your website doesn't
A traveller asks: “Where should we stay near Limassol Old Town and the waterfront?” The assistant suggests your hotel, but its displayed sources are third-party sites.
If AI can recommend the hotel, why is another website representing it? Outside OpenAI or Google, we cannot see exactly why one page was chosen for one answer. Their systems are proprietary. We can measure what appears.
Four events that are easy to confuse
An answer shows the first two, not the last two. Vektos measures recommendation and source representation separately, never presenting references as traffic or bookings. A hotel can be recommended often yet rarely represented by its own website.
Why AI answers show sources
Modern assistants can look things up while answering, rather than relying only on training. OpenAI says ChatGPT may automatically search for current information and include citations.1 Its search uses third-party search providers and content supplied by partners.2
For developers using Gemini, Google describes a process in which the model decides whether search would help, runs searches, combines results and returns citations.3 Google also describes Gemini trip planning drawing real-time locations, reviews and prices from Maps, Flights and Hotels.4
Your hotel exists in more places than your website
A traveller sees one hotel; the web holds many descriptions of it.
These sources overlap in name, location, rooms, amenities, guest suitability, policies, reviews and sometimes prices. Other public material can include forums, directories, partner pages, social posts and videos.
Different products and questions can draw on different parts of this picture. Booking platforms publish detailed, structured pages for many hotels, making their frequent appearance unsurprising.
So why might Booking.com appear?
An OTA page may appear among search results or contain the detail the traveller asked about. Another engine, question or day might surface the official website instead. These are possibilities, not proof that “Booking.com outranks us” or “ChatGPT prefers OTAs.”
- What we know
- Booking.com appeared as a displayed source in the measured answer.
- What we do not know
- Why the proprietary AI system selected that individual source or whether it caused the hotel recommendation.
What a source mix can look like
Measuring many answers turns individual screenshots into patterns. The historical Cyprus study recorded this distribution of source references.5
- Hotel websites
- 3.4%
- OTA platforms
- 44.9%
- Other sources
- 51.7%
Historical Cyprus sample
Recorded references — not bookings, clicks or traffic
These are historical sample-level findings, not a current score for an individual hotel. OTA references were more frequent than hotel-site references; the distribution does not explain source selection or prove a booking outcome.
Why this matters for the direct channel
Distribution teams already check direct-rate visibility, OTA content accuracy and public profiles. AI answers are another place where the hotel is described.
If third-party sites regularly represent a recommended hotel more than its own website, the direct channel may be underrepresented while travellers decide. This is a consideration risk, not proof of a lost direct booking. When the hotel or its direct site is underrepresented while alternatives are present, it may lose consideration before the traveller reaches a booking platform or hotel website.
Booking platforms remain distribution partners. The question is whether the hotel's own channel is present too, before the guest reaches a booking site.
Do ChatGPT and Gemini use the same sources?
Not necessarily. They are separate products with separate search systems and data connections. OpenAI describes search providers and partner content;2 Google's travel planning connects to Maps, Hotels and Flights.4 Measure each engine separately, across questions and dates.
What a hotel can actually measure
Record what happened consistently rather than trying to reverse-engineer the systems.
- Was the hotel recommended?Presented as an option, not mentioned in passing.
- Was the official website referenced?How often, and for which questions?
- Which other sources appeared?Booking platforms, reviews, destination sites and travel media.
- Did ChatGPT and Gemini differ?One may reference your site while the other uses third parties.
- Does it change by question?“Beachfront hotels” and “near the old town” can show different patterns.
Repeat the same questions later to see whether the pattern holds. A single screenshot offers one observation; a consistent question set provides a pattern that a hotel team can investigate and re-measure over time.
What to look at if your website is rarely referenced
Do not assume ChatGPT dislikes the site. Check what is actually public:
- Are core hotel facts clear on the official website?
- Are room types, amenities and policies current?
- Are location and nearby places clearly described?
- Are intended guest types—families, couples or business travellers—clear?
- Do OTA and review profiles match official information?
- Are recurring third-party sources accurate?
Your web team or agency can act on findings, then measure again. This is a process of measurement, investigation and action, not a guarantee that changing content will change future answers.
What a source reference does not prove
A website appearing as a source does not prove:
- that it caused the hotel recommendation;
- that the traveller clicked it;
- that the booking happened there;
- that the hotel lost revenue;
- that adding website content guarantees a future reference;
- that one answer is a stable ranking.
AI answers and sources change. Patterns across repeated measurements matter more than one screenshot.
The real question for hotel management
The useful question is not “How do we make ChatGPT cite us?” It is:
When AI helps a traveller consider hotels, how is our property represented, and is our direct channel there?
Unlike a secret algorithm, that is something you can measure.