This paper presents a mixed methodology for architectural heritage that combines UAV-based photogrammetry, large-language AI, and speculative storytelling to generate polyvocal narratives. Modern remote sensing (photogrammetry, LiDAR) provides precise 3D models of buildings, but these rich spatial datasets capture only physical form. Important intangible aspects (e.g.memories, local myths, and alternate histories) often remain monological and marginalized in official or authorized heritage discourse. We argue that integrating generative AI voices as “characters” (e.g. the building itself, materials, inhabitants) can amplify diverse perspectives and imagine new contexts. In our experiment, drone imagery is processed into 3D models (photogrammetric meshes) which become the backdrop for AI-generated narratives. Distinct AI “speakers” are seeded with site-specific facts and creative prompts to produce speculative accounts. These narratives, delivered as text-to-speech voices, interweave factual detail with imaginative reflection, creating a layered storytelling experience. We present a case study (McMahon Hall, Seattle, USA) to illustrate the workflow and results. The discussion highlights how this polyvocal, machine-augmented approach transforms heritage epistemology and how it can extend heritage documentation beyond objective recording into the realm of what-ifs, challenging singular histories and democratizing cultural memory.