COVID-19 YouTube Videos in the Late Pandemic: How Faces Detected on Thumbnails and Alignment of Thumbnails with Titles and Descriptions Relate to Engagement?

This work presents a large-scale analysis of how thumbnails of COVID-19-related YouTube videos depicted people, how these depictions varied across content categories and time, the alignment of these thumbnails with titles and descriptions across content categories, and how these factors drove engagement during the late-pandemic period. A dataset of 9,325 YouTube videos published between January 2023 and October 2024 was used for this study. A total of 17,546 faces were detected in the thumbnails of these videos using YuNet. Thereafter, the quality of these faces, estimated age, and age ranges were computed using FairFace and SwinFace, and the alignment between thumbnails and the titles and descriptions was investigated using the CLIP model openai/clip-vit-large-patch14. The results showed that conversational and creator-centric categories, such as People and Blogs and Entertainment, predominantly depicted younger adults, whereas Education, Science and Technology, and News and Politics were more likely to feature middle-aged adults, with these patterns shifting gradually across publication quarters. The engagement analyses, conducted after removing high-engagement outliers, indicated that thumbnails containing at least one face achieved higher median views in several categories, and that the relationship between the number of people shown and engagement varied by topic. The CLIP-based alignment analysis revealed patterns of variation in how closely thumbnails aligned with their title and description, with informational categories and thumbnails dominated by younger adults showing higher alignment. These findings demonstrate how thumbnail characteristics and their alignment with video titles and descriptions influenced the reach and engagement patterns of COVID-19-related videos on YouTube in the late pandemic period.