AI Thumbnails on YouTube for MSME Brands: A Study on GETI Media SME Live

Abstract

This study evaluates the effectiveness of AI-generated or AI-assisted thumbnails in enhancing content performance and reinforcing visual branding on the educational YouTube channel "GETI Media – UKM Live." The evaluation focuses on two key YouTube analytics metrics: Impressions Click-Through Rate (CTR), which serves as an indicator of visual appeal and its ability to drive clicks, and Average View Duration (AVD), used as a proxy for viewer engagement and retention. The research design employs an A/B experimental approach involving a minimum of 12 newly published videos over a three-week period. Each video is presented with two thumbnail conditions: Condition A (manually designed by a human designer) and Condition B (AI-generated or assisted using documented prompts). Controls are applied to variables such as video title, description, release time, tags, and duration to ensure consistency across conditions. In addition, a visual content analysis of the thumbnails is conducted to map elements that may moderate performance outcomes. These elements include: presence and expression of faces, clarity of logos or wordmarks, color contrast, information density, presence of visual call-to-action (CTA), brand identity consistency, composition, and gaze direction. Data analysis involves paired t-tests or Wilcoxon signed-rank tests to compare CTR and AVD across conditions, and logistic or linear regression models to assess the influence of AI-generated thumbnails and visual moderators, while controlling for covariates such as topic, video duration, impressions, and release timing. The study aims to provide empirical evidence and practical guidance for MSME (UMKM) practitioners and educational content creators in optimizing thumbnails while maintaining ethical standards and brand identity alignment. The findings are also contextualized within YouTube’s "Test & Compare Thumbnails" feature and the potential for cross-lingual visual localization.