TikTok Recommendation Algorithm: A UX-Focused Case Study of the 'For You' Feed

AI · 7 min read

We analyzed feed patterns across accounts and measured how subtle interface changes affect dwell time. TikTok combines dense user signals — watch time, replays, completions — but the UI hides the feedback loop that trains the model. This opaqueness maintains engagement but erodes user trust when content becomes repetitive or toxic.

Our teardown shows that small transparency cues (e.g., ‘Because you watched X’ badges with dismiss options) can give users agency while preserving retention. We prototyped a micro-interaction that lets users mark a video as “less like this,” and observed a reduction in similar recommendations within ten sessions for test accounts. The interaction is lightweight enough to avoid interrupting habitual scrolling.

We also explored the cost of personalization on diversity: heavy optimization for maximized watch time funnels users into narrower taste corridors. Solutions like scheduled “explore bursts” and UI-level serendipity toggles would let users modulate between comfort and discovery, aligning UX with long-term satisfaction rather than short-term engagement metrics.