YouTube channels communicate their identity through repeated use of titles, descriptions, transcripts, and categories. While prior research has focused on engagement and cross-feature alignment, less attention has been given to how consistently these elements are used within a channel. This paper presents a content-based framework for characterizing channels based on semantic similarity across videos within each feature. Using a dataset of 150 channels and 157,235 videos, we computed inner similarity scores for each content field and analyzed six pairwise combinations. Five unsupervised clustering algorithms were applied to each combination, and stable groupings were identified through majority voting. The resulting clusters revealed three characterizations: channels with low internal consistency, those with fixed metadata and varied content, and channels with strong structural coherence. These patterns highlight differences in editorial strategy and presentation. The method is scalable, language-agnostic, and label-free, offering a new perspective on how channels maintain or vary messaging across content.