Ai-generated Podcasts As A Novel Educational Tool In Plastic Surgery: Student Perceptions And Comparative Analysis

Abstract

PURPOSE: AI is increasingly used to deliver and enhance education across formats. AI-generated podcasts offer one such opportunity; however, their educational quality, clarity, tone, and trustworthiness in plastic surgery education remain largely unexamined. To address this gap, we assessed medical students perceptions of AI-generated podcasts derived from PRS articles. METHODS: Three craniofacial surgery articles from Plastic and Reconstructive Surgery were converted into podcasts using AI tools: Google NotebookLM (A), Wondercraft (B), and Jellypod (C). Eight medical students read the articles before a journal club session, where all podcasts were played sequentially. After each, participants completed a 17-item, 4-point Likert survey adapted from the SSEPQ, QAEP, and Desmedt et al., evaluating four domains: Design & Structure, Content Accuracy, Tone & Credibility, and Learning/Engagement. Descriptive statistics and paired Wilcoxon signed-rank tests with Bonferroni correction were performed in RStudio 4.5.1. RESULTS: Podcast A achieved the highest ratings across all evaluated domains. The majority of responses rated it favorably in Design & Structure (91.7%, 44/48 ≥ 3), Content Accuracy (100%, 32/32 ≥ 3), Tone & Credibility (91.7%, 22/24 ≥ 3), and Learning Value (96.9%, 31/32 ≥ 3), with few low ratings (≤ 2) observed. In contrast, Podcast B demonstrated lower performance across domains, particularly in Learning Value, with only 62.5% (20/32) of responses ≥ 3 and 37.5% (12/32) ≤ 2. Tone & Credibility (79.2%, 19/24 ≥ 3) and Content Accuracy (84.4%, 27/32 ≥ 3) were also comparatively reduced. Podcast C received high ratings for Content Accuracy (100%, 32/32 ≥ 3) and Design & Structure (87.5%, 42/48 ≥ 3), moderate ratings for Learning Value (84.4%, 27/32 ≥ 3), and greater variability in Tone & Credibility (75.0%, 18/24 ≥ 3; 25.0%, 6/24 ≤ 2). At the item level, 87.5% (7/8) of participants found Podcast A enjoyable to listen to, and all (8/8) indicated they would recommend it to others. Fewer participants endorsed Podcast B (37.5%, 3/8 would recommend), consistent with lower ratings in tone, engagement, and perceived credibility. Podcast C was generally well received, although 25% (2/8) of participants reported concerns regarding tone. Median item scores were 3 across all podcasts and items. Wilcoxon signed-rank tests with Bonferroni correction demonstrated no statistically significant differences between podcasts (A vs B: p = 0.773; A vs C: p = 0.346; B vs C: p = 1.000; all adjusted p ≥ 1.000). CONCLUSION: AI-generated podcasts were well received for accuracy and educational value, suggesting their potential as effective adjuncts in plastic surgery education. Further research should explore how tone and design affect learner engagement. *Source: https://ps-rc.org/meeting/Program/2026/56.cgi*

Ähnliche Arbeiten