[Aug 25] GPT-4 vs GPT-5 for 5 Global cities: Booking.com and OTA Links in AI Hotel Recommendations

From 17% OTA links with GPT-4 to just 1% with GPT-5 β€” a seismic shift in AI travel search.

β€’β€’4 min read

AI and SEO expert at the forefront of AI Search. He analyses models daily and runs hospitality-focused experiments on a database of over 1M prompts, citations and mentions.

Introduction

Are AI search engines moving away from OTAs like Booking.com? Our latest analysis suggests a dramatic yes. Comparing GPT-4o and GPT-5, we measured how often each model included OTA links versus direct links in response to 3,000 traveler queries.

Methodology

We processed 3,000 real traveller prompts across five major global cities β€” New York, London, Hong Kong, Tokyo, and SΓ£o Paulo β€” and three traveller personas: Business travellers, Couples, Families with children. For each AI response, we classified links as: OTA (Booking.com, Hotels.com, Expedia, etc.) or Direct (official hotel websites).

Donut chart comparing OTA vs direct hotel links in GPT-4o and GPT-5 responses.

Key Findings

  • With GPT-4o, OTA links accounted for 17% of all recommendations
  • With GPT-5, OTA links dropped to just 1% (!!)

This shift is consistent across all cities and traveller personas.

Conclusion

The data points to a major distribution shift: GPT-5 virtually eliminates OTA links from its recommendations. For hotels, this is an opportunity to capture more direct bookings β€” but only if they ensure their properties are visible, relevant, and optimised for AI-driven search.

Nicolas Sitter

Nicolas Sitter

AI and SEO expert at the forefront of AI Search. He analyses models daily and runs hospitality-focused experiments on a database of over 1M prompts, citations and mentions.

Co-founder of Hotelrank