Comparison of NotebookLM-generated podcasts and article reading for radiology learning

Though podcasts are becoming increasingly popular in medical education due to their convenience, portability, and amenability to multitasking, their role in radiology education remains unclear. The emergence of artificial intelligence tools such as Google's NotebookLM, which can generate structured podcasts from peer-reviewed journal articles, presents a novel opportunity to integrate podcast-based learning into radiology education. We conducted an institutional initiative comparing resident learning using NotebookLM-generated podcasts to reading radiology research articles on knowledge acquisition and learner satisfaction. Reading produced more consistent knowledge gains, whereas gains with podcasts were more heterogeneous. Despite this, residents perceived the podcasts as more engaging, convenient, and less cognitively demanding. Qualitative feedback indicated that while podcasts show promise as accessible learning tools, they are currently constrained by limited technical detail, repetition, and the absence of visual content. In contrast, articles were described as comprehensive and rich in visuals, but also lengthy and dense. Our findings suggest that NotebookLM-generated podcasts are best positioned as complementary resources rather than replacements for reading, providing a brief overview of an article or reviewing previously studied material. Educators should consider verifying generated content for accuracy and pairing podcasts with original articles when visual or technical details are important.