The Algorithmic Grand Tour

Abstract

Preregistered audit of how the models behind three generative AI assistants (OpenAI GPT-6 Astra, Google Gemini 3.1 Pro, Anthropic Claude Opus 5.5), accessed through their APIs, recommend cultural itineraries in Greece. One fixed trip request is sent for each of the eight visitor profiles of the ACUX typology (Konstantakis et al., 2022, Digital 2(3), 365–378) and a no-profile baseline, in three trip scenarios (Greece 7 days; Athens and Attica 3 days; Heraklion, Crete 4 days), three times each: 243 responses. We measure how strongly recommendations concentrate on the most visited sites, which heritage remains invisible, whether itineraries adapt to the visitor profile, and how the models differ. No human participants.