From Geospatial Data to Narrative: A GIS-LLM Pipeline for Generating Personalised Outdoor Route Descriptions

With the growing availability of detailed geospatial data and the rise of generative AI, there is increasing potential to enhance how route-based information is communicated to end users. Current cartographic practices rarely incorporate engaging narrative descriptions, and while generative AI can address this gap, factual inaccuracies ("hallucinations") present significant challenges. These hallucinations can dangerously mischaracterise terrain difficulty and compromise walkers' safety (for instance, mistakenly labelling steep or uneven paths as suitable for families with young children or individuals with limited mobility). This work presents a GIS-LLM pipeline that generates human-centred walking routedescriptions while highlighting the importance of accuracy and safety.