<![CDATA[Video content is increasingly central to distance and mobile learning, yet the volume of instructional media available to learners creates a pressing need for tools that can efficiently distill key content without compromising pedagogical integrity. This study comparatively evaluates three AI-based video summarization tools OpenAI, SumTube, and NoteGPT applied to authentic educational media from UT Radio Mobile at Universitas Terbuka. Eleven videos across two content categories were analyzed: promotional programs (Seputar UT) and module-based instructional content (Tutorial Radio). Using a comparative evaluative design, summary outputs were assessed against four integrated criteria accuracy, suitability, clarity, and processing time by expert validators, with inter-rater consistency reported across content categories. Results indicate that OpenAI achieved the highest accuracy for promotional content (93.3%) through narrative-preserving abstraction, while NoteGPT demonstrated stronger performance on instructional content (85.7%) but exhibited systematic semantic drift including terminological substitution, logical inversion, and topic conflation with meaningful consequences for novice learners. SumTube consistently delivered the fastest processing times (~4 minutes), making it suitable for time-constrained mobile learning contexts, though its extractive architecture rendered outputs susceptible to non-instructional content inclusion. Beyond tool benchmarking, the study demonstrates that summary quality has direct implications for cognitive load, knowledge retention, and learner engagement, and that prompt engineering functions as a form of pedagogical mediation that shapes the instructional alignment of AI-generated outputs. Findings support a hybrid, content-sensitive summarization model and offer theoretically grounded guidance for integrating AI summarization tools into distance and mobile education platforms.]]>
