Abstract:
According to this thesis, users' perceptions and interpretations of AI-powered recommendation systems in subscription platforms influence their judgments about awareness, choice, trust, and continuity. Fifteen active users from Generation Z, Millennials, and Generation X participated in three generationally divided focus groups using an interpretivist qualitative design. Reflexive theme analysis was used to analyze the data, and English translations of the exact quotes were used as support. Results clearly indicate domain imbalance. Short trials, simple interruption, and quick feedback loops make recommendations reliable starting places for learning in musical situations. Users apply more scrutiny to long-form videos due to the higher time cost and larger preference salience; they frequently replace platform feeds with external anchors like IMDb and personal recommendations. Rather than relying on technical knowledge of algorithms, perceived personalization is evaluated based on pragmatic indicators of control, fit, and recentness. Trust is based on shared values of competence, honesty, and kindness. Although competence can be recognized, trust is nonetheless damaged when home screens are interpreted as being driven by promotions or trends and when there is insufficient control to adjust the feed. Acceptance can be restored with brief justifications like "because you watched" and simple controls to muffle a theme. Although behavioural influence is genuine during sessions, it is limited in terms of subscription continuation, which is assessed based on factors including price, catalogue scope, free time, and household requirements. Cohort differences are not absolute; they are structured. While Generation Z is wary of trend-heavy video streams, they incorporate recommendations into their everyday routines. Fit-first pragmatism and opposition to perceived franchise pushing are two opposing viewpoints among millennials. The most cautious generation, Generation X, depends on outside validators and cancels when value drops. While recognizing the limitations of the qualitative, region-specific corpus and suggesting mixed method routes for cumulative research, the paper presents design implications that value visible control, succinct explanations, and calibrated diversity.
