Artificial intelligence-generated podcasts open new doors to make science accessible: a mixed-method evaluation of quality and the potential for science communication.

Aims: Science podcasts have proved to be valuable mediums for medical education and science dissemination. Tools adopting rapidly evolving technologies such as generative artificial intelligence (GenAI) now enable us to create podcasts in a matter of minutes (e.g. NotebookLM, Jellypod). However, GenAI entails challenges, such as hallucinations, which could compromise the trustworthiness of generated content. Therefore, this study aimed to explore the quality of AI-generated podcasts and their potential for science communication.

Methods and results: We conducted a mixed-method evaluation of 10 AI-generated podcasts for articles published in the European Journal of Cardiovascular Nursing. Participants were asked to complete a questionnaire and were invited for a video interview. They were not informed of the AI-nature of the podcast prior to evaluation. Only half of them were able to identify this aspect. The fact that the podcast was able to summarize key findings in an easily understandable and engaging manner was found to be a great asset. However, participants also indicated that the American style of the podcast took away from its credibility. Moreover, some podcasts contained inaccuracies, incorrect use of medical terms and mispronunciations, thereby compromising trustworthiness. Podcasts were found to be most appropriate for patients and the public but could be useful for researchers and healthcare professionals as well if they were tailored accordingly. Rigorous evaluation and transparency about the AI-generated nature of the podcast, referencing the original article and author acknowledgement were recommended.

Conclusion: AI-generated podcasts could be relevant additions to scientific journal articles and valuable alternatives for traditional science podcasts.