700 Section - AI vs. Human: Can ChatGPT Find My Accessible Room?

Wednesday, May 20, 2026

6:00 PM - 7:30 PM PT

Co-Author/Coautor(s)

CL

Carla Lee Kim, MD

Research Lead
Philippine American Physiatrist Association (PAPA)
Santo Domingo, Distrito Nacional, Dominican Republic

Presenting Author/Autor expositor(s)

Niña R. Carmela Tamayo, DO MS MPH (she/her/hers)

DO MS MPH
Prime Directives PM&R, Cleveland, OH and PAPA
Lakewood, Ohio, United States

Co-Author/Coautor(s)

KP

Kurt Panganiban, BA

Medical Student
Western University of Health Sciences College of Osteopathic Medicine of the Pacific
Oxnard, California, United States
KF

Kai O. Francisco, BA

Co-Author
Michigan State University College of Osteopathic Medicine
East Lansing, Michigan, United States
AN

Amogh Nagol, BS

Co-author
Drexel Medical School
Wyomissing, Pennsylvania, United States

Objectives : Accessing reliable data on hotel accessibility remains a persistent challenge for travelers with disabilities. As artificial intelligence becomes more prevalent, an important question arises, can these systems provide accurate and reliable information to meaningfully support disabled travelers? Tamayo et al. (2019) demonstrated that Hilton hotels offered the highest number of accessibility features. Following this finding, the present study evaluated whether ChatGPT could accurately identify said brand and affiliated hotels in Manhattan, New York, and extract detailed wheelchair-accessibility information compared with manual verification.

Design: We developed a standardized prompt through five iterations to direct ChatGPT to generate a comprehensive dataset of Hilton Brand, subsidiaries, and affiliated hotels in Manhattan, New York. The prompt instructed the model to classify hotels by neighborhood and star rating while simultaneously extracting accessibility features relevant to wheelchair users, including door width, step-free routes, ramps, roll-in showers, grab bars, elevators, parking, and bedroom/bathroom turning radius. The AI-generated dataset was then compared against manually collected data to evaluate accuracy and completeness. Invalid or missing information was verified through direct property websites.

Results: ChatGPT identified 30 Hilton-affiliated properties, whereas manual review found 48. Several hotels listed by ChatGPT were non-existent or inactive. Contact information, including phone numbers and email addresses, was frequently incorrect. Among accessibility features, ChatGPT inconsistently reported the presence of wide doors and grab bars, while manual review provided more reliable and complete data.

Conclusion: Despite specific prompts and explicit criteria instructions, ChatGPT consistently produced incomplete and inaccurate accessibility data. Manual verification showed to be more consistent as well as more accurate/comprehensive. While AI systems show potential to streamline data gathering, they are not yet reliable enough for accessibility research or for travelers with disabilities who rely on precise information. Manual review of official sources remains the recommended method until further developments in AI are made.