Spotify is expanding its AI-powered personalization capabilities by trialing "Prompt Playlists," a new feature launched on January 28, 2026. This system allows users to generate custom music playlists by inputting text-based descriptions of desired moods or situations. The core methodology leverages a sophisticated artificial intelligence framework that processes natural language prompts from users. This input is then cross-referenced with two primary data sources: the user's extensive historical listening data (user profile and preferences) and broader music trend information (catalog popularity, genre trends, new releases).

Technically, the process can be conceptualized as follows:

1. Natural Language Understanding (NLU): User text prompts are parsed and understood to extract underlying musical attributes, emotional connotations, and contextual cues.
2. Contextualized Recommendation Engine: The NLU output acts as a dynamic query constraint for Spotify's recommendation engine. This engine, often based on a blend of collaborative filtering (e.g., matrix factorization techniques like Singular Value Decomposition, SVD, or Alternating Least Squares, ALS, over user-item interaction matrices) and content-based filtering (analyzing audio features, metadata, and genre tags of songs), then identifies candidate tracks.
3. Personalized Filtering: The candidate tracks are then filtered and ranked based on the user's explicit listening history and implicit preferences. This involves calculating similarity scores between candidate tracks and the user's past consumption, potentially using embedding vectors for both users and tracks (e.g., derived from deep learning models like Word2Vec for music or more complex sequence models).
4. Constraint Application: A unique aspect is the ability for users to impose additional, more granular rules, such as "based on artists listened to in the last 5 years but include unlistened songs." This suggests a rule-based system or an optimization layer applied post-recommendation generation, further refining the playlist output. This can be seen as an optimization problem where the objective is to maximize relevance to the prompt and user preferences while satisfying user-defined constraints.

The feature was initially tested in December 2025 with Premium users in New Zealand. A beta rollout subsequently began for Premium users in the United States and Canada on January 22, 2026. This functionality aims to enhance user control over music recommendations, moving beyond passive auto-recommendations. Users can set specific conditions to refine the generated playlists. Playlists can be configured to refresh daily or weekly, preventing content stagnation. Mobile Premium users can access the feature via a "Create" button within their library and can modify prompts even after playlist generation. The system also provides a concise one-line note explaining the inclusion of each song, with an option to toggle these notes. As a beta feature, it is subject to potential usage limits, and deleted prompt playlists are irrecoverable. Currently, the service is exclusively available in English.