The fallacy of AI-generated podcasts: the case of Google NotebookLM

This paper critically examines science communication discourse produced through AI-generated podcasts, created via Google NotebookLM. We propose a rhetorical-argumentative framework to verify whether podcasts mediating scientific reports propagate quality discourse or mis−/dis-information. We apply this framework to the qualitative analysis of a corpus of podcasts stemming from scientific reports about communication and media research. The study reveals that podcasts produced via the ‘Audio Overview’ feature of Google NotebookLM, though engaging, trivialize content, propagate biases, and draw unwarranted inferences that misrepresent authors' intentions. The integration of dialectical and rhetorical components within our discourse analysis surfaced the presence of fallacies and misleading ways of packaging and presenting information. This misrepresentation risks undermining research integrity, as authors are not guaranteed the opportunity to verify the accuracy of their work before dissemination, and it spreads misinformation, instead of ensuring quality science communication to inform policymaking.