This research investigates the application of artificial intelligence (AI) to the study and conservation of the Sheikh Lotfollah Mosque, one of the most celebrated monuments of Safavid architecture in Isfahan, Iran. Using a multi-layered methodology, the study evaluated the performance of AI systems in five key domains: crack detection, geometric motif reconstruction, generative modeling, color restoration, and image enhancement. Results demonstrated that convolutional neural networks achieved recall rates of 96% in identifying micro-cracks, thereby outperforming traditional manual inspection. In motif reconstruction, polygons were restored with structural similarity (SSIM) values above 0.89, while arabesques and calligraphy remained more challenging. Generative adversarial networks produced geometrically sharp motifs, whereas diffusion models excelled in perceptual realism, suggesting hybrid potential for conservation practice. In color restoration, ΔE values remained below or near the perceptual threshold of 3.0, confirming chromatic authenticity, while image enhancement improved PSNR by 14.2 dB, revealing micro-cracks and glaze details previously undetectable. Symmetry and tessellation analysis validated that AI internalized the mathematical rules of Safavid ornament, although its tendency toward perfection underscored the importance of human oversight in preserving authenticity. The findings establish AI as a reliable diagnostic and reconstructive partner while also raising questions of authenticity, authorship, and cultural stewardship. The study concludes that AI should be framed as a digital apprentice in heritage science, capable of extending the life of Iranian architectural traditions in both material and computational domains.