AI search is sending travellers to official websites
Ask a traditional search engine for a hotel and you mostly end up on a booking site. Ask an AI assistant and you mostly end up on the hotel's own website. That is the finding of a Morgan Stanley analysis published on 2 November. Across hotel searches in eight US destinations, traditional search sent travellers to official hotel websites 13% of the time. AI-powered search did so 97% of the time. Morgan Stanley says it saw similar effects in Europe.
The research covers hotels, not tours. But the same pattern is showing up elsewhere in AI commerce, and the lesson for tour and activity suppliers is the same: AI assistants prefer the official source when they can find one and trust it.
What did the research find?
| Finding | Figure | Source |
|---|---|---|
| Hotel searches sent to official websites by traditional search | 13% | Morgan Stanley, 2 Nov 2026 |
| Hotel searches sent to official websites by AI search | 97% | Morgan Stanley, 2 Nov 2026 |
| AI links to official sites when the traveller is in the hotel's loyalty programme | 100% | Morgan Stanley, 2 Nov 2026 |
| ChatGPT "Shop at" buttons pointing to the brand's own store, when the brand's page is not a cited source | 6% | GEOly, 27 Oct 2026 |
| The same, when the brand's page is a cited source | 34% | GEOly, 27 Oct 2026 |
The second pair of figures comes from retail. In October, ChatGPT began adding a "Shop at" button under products it recommends. According to GEOly's analysis of WISLR data, 56% of those buttons pointed to big-box retailers and about 5% to the brand's own store. When the brand's own product page was one of the sources the answer cited, the brand's share rose from 6% to 34%.
Two very different markets point the same way. The model sends the buyer to the official source when that source is part of the answer.
Why does it matter if the click goes to your own site?
Because the margin difference is large. Morgan Stanley puts commission on third-party hotel bookings at 10–20% for branded hotels and 20–30% for independents, against roughly 5% for a direct booking on the hotel's own website. Tour and activity suppliers know those numbers well. We compared what suppliers keep on Viator, GetYourGuide and Klook in September.
A direct booking also keeps the customer. The supplier gets the email address, the chance of a review on its own channels and the repeat visit. A marketplace booking keeps all of that with the marketplace.
Does the OTA lose out completely?
No, and the research shows where it wins. When Morgan Stanley told the AI agent that the traveller belonged to an online travel agency's loyalty programme, the results shifted entirely to that booking platform. The agent followed the relationship it was told about.
That is the risk for tour and activity suppliers. Big hotel groups have loyalty schemes with millions of members to lean on. Most tour suppliers do not. If an assistant knows a traveller uses a particular marketplace, it has a reason to send them there. A supplier without that relationship has to win on what the model can read: clear, current, bookable information on its own site.
Is AI traffic big enough to matter yet?
Not in volume. Booking Holdings said on 3 November that referrals from large language models are still significantly less than 1% of Booking.com's room nights, though they have accelerated in recent weeks.
The quality of that traffic is already high, though. Adobe found that AI traffic to US travel sites rose 119% year on year in July 2026 and now converts only 1% below other traffic, against 47% below a year earlier. Travellers arriving from AI answers are ready to book.
The habits being set now will decide who gets that traffic as it grows. Google is moving the same way. On 15 October it said it is testing a dedicated section for official websites at the top of some AI Mode and AI Overview responses, alongside more inline links.
Why isn't being mentioned enough?
Because a mention without a link sends the booking somewhere else. Semrush's AI Visibility Index found that fewer than half of the brands AI assistants mention are among the domains they cite. A model can name your walking tour and still link to a marketplace listing, because that is the page it could read and verify.
The GEOly data shows the same gap. Being recommended is one step. Being the cited source is what moves the click to your own store.
What makes a supplier's site the one AI picks?
The model has to be able to confirm three things from your own domain:
- What the product is. Name, duration, location, inclusions, languages and meeting point, in structured data as well as page copy.
- That it is real and current. Live prices and availability, cancellation terms and recent reviews the model can check.
- That it can be booked there. A booking flow on your site that works for a visitor who arrives knowing what they want.
Many tour and activity suppliers' websites fall short on at least one of these. Prices sit in images or PDFs. Availability lives only in the booking system. The checkout is a redirect to a third-party page. In each case, the marketplace listing is easier for the model to read and trust, so that is the page it cites.
A quick check: ask ChatGPT, Gemini and Perplexity for your main tour by name and by category, such as "best food tour in Lisbon". Note which pages they link to. If the answer is a marketplace rather than your site, the model could not use your site as the source. Our guide to auditing what AI assistants say about your tours sets out the full method.
Where Tixxly fits
Tixxly is AI infrastructure for tour and activity suppliers. It is not an OTA and charges zero commission. We publish suppliers' products as structured, AI-readable storefronts under their own brand and domain, with live availability and embedded checkout where the supplier is the merchant of record. We measure how often each supplier is mentioned and cited across the major AI assistants. The aim is simple: when AI looks for the official source, it finds yours, and the booking comes to you.