Structured Prompt Engineering for AIGenerated Cultural Tourism Narratives: A Comparative Study of Ulysses and Spirited Away

Abstract

Narrative is a vital medium in cultural tourism. It connects visitors with place, memory, and identity, moving beyond factual information to create meaning. With the rise of generative AI, large language models (LLMs) can produce digital travel narratives. However, their outputs often lack coherence, cultural alignment, and personalization. This study examines how structured prompt engineering can guide LLMs to generate more coherent and culturally grounded narratives. Using GPT-4o as the model, the research applied a three-stage prompt design. V1 provided only literary framing. V2 added geographic anchoring. V3 introduced personalization through parallel side notes. Two contrasting cases were used: Ulysses by James Joyce and Spirited Away by Hayao Miyazaki. Twelve narratives were generated and evaluated on five dimensions: narrative coherence, cultural alignment, personalization, functional operability, and interpretive depth. The results show that prompts act as decisive controls rather than neutral instructions. V1 produced fragmented but stylistically consistent stories. V2 turned locations into narrative scaffolds, improving coherence and usability. V3 added personalization while preserving stability by keeping it in side notes. Cross-case comparison revealed genre sensitivity. Spirited Away, with its episodic cinematic structure, gained strongly from geographic scaffolding. Ulysses was a special case. Its narrative already followed a geographic path but with complex style and symbolic layering. For this reason, the gains were smaller and appeared mainly in cultural detailing and stylistic refinement. This study contributes theoretically by reframingprompts as mechanisms of narrative control. It contributes methodologically by introducing layered prompt design and a combined evaluation framework. It contributes practically by offering strategies for cultural tourism and digital heritage. At the same time, it highlights risks of formulaic reproduction and cultural flattening. The study positions LLMs not as autonomous storytellers but as collaborative tools. With structured prompts and expert curation, they can enrich cultural storytelling and expand access to heritage.