Designed to land

Depressive and demotivational videos circulate widely on Instagram Reels and TikTok, platforms whose audiences skew young, and depression rates in that age group have been rising. A recent meta-analysis of 58 studies covering 96,676 participants reports an association between short-form video use and poorer mental health (Tang et al., 2026), though that evidence is largely cross-sectional and the direction of causation remains unsettled. What is missing from this picture is any account of the content itself. This paper analyses 129 depression-related Instagram Reels, assembled over five weeks from a single account by using the Like function both to archive videos and to signal interest to the recommender; 500 liked videos were screened to reach 154 candidates and 129 confirmed. Each was processed through a two-tier pipeline combining speech transcription, scene-change frame sampling, acoustic feature extraction, and structured description by a local vision-language model. The videos share a narrow set of construction techniques regardless of subject: a solitary figure or vehicle in an empty landscape, one line of text held unchanged for the full duration, direct address to the viewer, low-arousal music, and almost no narrative resolution. Over half the corpus uses one of two interchangeable visual setups, 40% contains no speech at all, and the median video runs 12.2 seconds. Repetition is literal rather than thematic: one caption appears on three otherwise unrelated videos, and footage of the same actor, taken from existing films, carries two different captions. These techniques resemble devices that experimental research links to rumination and to failed mood repair. A content analysis cannot establish an effect on viewers, and none is claimed.Data collection, pipeline development, coding, and analysis by the author. An AI assistant was used for drafting and copy-editing.