What Visitors Say and What Icons Show: Re-Examining a Cultural-Visitor Typology for Generative-AI Personalisation in Heritage Guides

Heritage guides often profile visitors with a typology and icons, while generative artificial intelligence (AI) now offers to infer a profile from a visitor’s own words. Whether either captures how visitors see themselves is rarely tested. In an exploratory survey, 37 cultural visitors, mostly in Greece, gave the same preferences in four ways (reading six unlabelled icons, choosing interest pictograms, choosing photographs and describing an ideal day) and rated themselves on the eight profiles of the ACUX typology. A locked large-language-model profiler (Gemini) scored the texts. Only two of six icons were read as intended by over two thirds of respondents, and misreadings were systematic. Visitors described themselves as mixtures of profiles and formed no distinct clusters. Beyond the profile shape shared by the whole sample, pictograms ranked first and the AI profiler last, but no difference survived multiple-comparison correction, and the AI’s agreement rose with text length. The AI best reproduced the average profile, consistently across runs. Over half the texts described how to visit, which no profile names. Visitors valued explanations more than they trusted AI suggestions. Heritage guides should label and test icons, keep profiles multi-label and correctable, and validate AI profilers beyond the average visitor.