ArrivlReached guest details
The official booking path reached the guest-information stage in our browser tests.
24 of 27 hotel testsARRIVL RESEARCH / NEW YORK CITYAI reached the booking form
AI reached the booking form
on most hotel sites we tested.
We tested official booking paths at New York City luxury hotels. Most reached the guest-details page, while date selection and access problems stopped others earlier.
OFFICIAL BOOKING PATHNew York City
REPLAYING STUDY RECORDS
89%reached guest details24/27
11%stopped earlier3/27
Explore simulated prompts during the 15-second pause. Tooltip recommendations are illustrative.
Homepage accessible without JavaScript
A simpler request could access fewer of the same hotel homepages.
17 of 27 hotel testsHomepage accessible in a headless browser
Enabling JavaScript changed access in some cases.
19 of 27 hotel testsThe browser changes
what an agent can do.
A browser reached further
The browser walkthrough reached guest details at 24 hotels. Simple requests and headless browsers could access fewer homepages in the same sample.
Dates and access can interrupt the booking
The Greenwich Hotel stopped at the booking interface. Trump International and The Plaza reached the date-query stage but did not reach guest details.
Recovery sometimes needs another route
The recorded tests include finding a hotel through the brand website and using a dated booking URL when the calendar did not work.
A city-wide recommendation study
The next study will add repeated traveler prompts, a consistent hotel register, recommendation frequency and source links. The existing booking-path tests provide the first findings.
Methods & evidence
Study scope and methodology
The sample contains 27 New York City hotels from the union of Forbes Travel Guide four- and five-star properties and AAA five-diamond hotels compiled on October 21, 2026. It is a selected luxury-hotel sample.
Each hotel was checked using a static request, a JavaScript-enabled headless browser and a Chrome walkthrough. The recorded results show 17/27 homepages accessible to static requests, 19/27 to headless browsers, and 24/27 walkthroughs reaching guest details on the official booking path. Each hotel was tested once in this dataset.
The walkthrough requested December 11–13, 2026, two adults and one room, with December 8–10 allowed when the initial stay was unavailable. The protocol permitted a dated-URL fallback after repeated calendar failure and recorded route recovery. Tests stopped before guest information, login, payment or reservation submission. No reservation was completed.
These are observed website-access and booking-path outcomes. They do not measure how often AI recommends these hotels, and do not establish a city-wide completion rate. Percentages are rounded.
Source: Arrivl hotel-agent tests, October 2026.