Being recommended and being easy to book are different requirements. Use this historical test to examine the steps between an AI recommendation and your own reservation process, rather than as a promise of what every agent will do today.
What we asked it to do
The original July 2025 test asked ChatGPT Agent to plan a Budapest weekend: find an Airbnb under €100 per day, suggest activities and include vegan restaurants. The task brought accommodation search and an itinerary into the same conversation.
Watch the original recording on YouTube ↗
Read the test summary
The task combined accommodation research, activities and vegan restaurant suggestions for a Budapest weekend. The original account describes the agent navigating an accommodation booking path and reaching a stage that required the user’s login and payment involvement. It did not establish a completed, paid reservation or a direct booking with a hotel.
What the test showed—and what it did not
The original account describes the agent researching options, assembling a plan and progressing into an Airbnb reservation flow. It paused when the traveller needed to take over for login and payment. That reduced some manual navigation in this particular task; it did not remove the traveller’s role in authorising a booking.
This was one observed journey. We did not benchmark reliability across a large sample, verify a completed paid stay, or demonstrate a hotel’s own direct-booking integration. An agent navigating a third-party website is also different from a supported connection between an AI platform and a hotel’s reservation system.
- ResearchInterpret the request and find relevant options.
- PrepareAssemble the itinerary and progress through a booking path.
- User takeoverCheck the details, provide required access and approve payment.
What should a tourism business check?
A recommendation is useful only if the proposed next step works. Pick a real guest task and check it end to end: can someone find the correct experience or room, understand the price and terms, and reach a functioning reservation step?
- Identify the seller. Is the route through your own channel or an intermediary? Who receives the reservation and handles the guest?
- Check the information that affects the choice. Dates, location, inclusions and cancellation conditions must be clear where the traveller makes the decision.
- Find the points of friction. Note where a guest must re-enter information, create an account, resolve an unclear option or ask for help.
- Measure the outcome. Separate a recommendation, a booking-page visit and a confirmed reservation.
For a destination, the corresponding question is which places reach the suggested itinerary at all. A polished booking path cannot compensate for an experience that the traveller never encounters.
How to read this in 2026
OpenAI’s documentation describes web search as a way to retrieve current information, and browser interaction can involve rendered pages. Behaviour depends on the product, task and available tools. Test your own booking path before making a commercial commitment. Web search documentation · Browser documentation.
For hotels considering an integration today, the commercial distinction is between appearing in an answer, sending a guest to a booking link and supporting a connected reservation. Our AI hotel booking guide explains those routes and the limits of the currently announced rollout.
Assess your own booking path
If you are considering an AI booking project, start with the reservation system, platform access and commercial route you actually control.
Explore AI booking readiness for hotels →For destination research, explore which places appeared in the Balaton study.
Editorial note: Original article and test by Bence Bukovec, published 25 July 2025 and previously updated 19 August 2025. This edition retains the original recording and separates its observations from current guidance.