AI-generated podcasts in higher education: an author-evaluated pilot study of automated audio summaries of scientific papers

Podcasts are increasingly used in higher education to support learning and engage students. Recent advances in artificial intelligence (AI) now allow automated podcast-style summaries of scientific articles. This pilot study examined the quality and didactic potential of such summaries. Thirty-three authors of peer-reviewed social science articles generated English-language audio summaries of their own papers with Google’s NotebookLM and evaluated them using standardised ratings and open-ended responses. Authors rated the summaries favourably: 91% reported that the podcasts accurately presented key content, and 94% rated the listening experience as engaging. However, only 42% agreed that the findings were presented in a balanced way. Perceived suitability was higher for non-scientific audiences (82%) than for scientific audiences (55%; V = 309, p = .01, rrb = .63). Authors were also more willing to use or recommend podcasts based on their own papers than on papers written by others (V = 163, p = .049, rrb = .41), suggesting that source proximity and opportunities for review matter. Qualitative responses highlighted the importance of editing functions, editorial control, and clear labelling. Overall, AI-generated podcast summaries show potential for higher education and science communication in the social sciences, provided that appropriate review and transparency mechanisms are in place.