Spotify
genAI voice DJ (synthetic voice plus generated commentary)
Spotify's AI DJ (launched 2023, 60+ markets) captures about 25% of listening time on usage days and saw engagement nearly double in a year, according to Spotify's press releases.
Key points
- genAI voice DJ that chains personalized tracks and comments on them out loud.
- Sonantic synthetic voice, editorial scripting scaled by generative AI.
- 25% of listening time captured on usage days, engagement nearly doubled in a year.
- Evidence B, confirmed status, figures from Spotify press releases (2023 and 2025).
Objective
Reduce the friction of music choice by giving each listener a voice guide that chains tracks together and explains its choices, to bring the experience closer to a personalized radio station and increase time spent in the app.
The deployment
The DJ combines three building blocks: Spotify personalization (what to play), commentary written by an editorial newsroom of music experts and then shaped by generative AI (what to say), and a realistic synthetic voice from Sonantic technology. The listener presses a button, the DJ starts a personalized stream and comments on artists and tracks out loud. Since 2025, the listener can also make voice requests (genre, mood, artist) to reorient the session in real time.
The case in action
Official videoMeet Spotify's A.I. DJ · voir sur YouTube
Results Proof B
Engagement figures communicated by Spotify in its own press releases (newsroom), with two measurement points at two dates (2023 beta, 2025 update). Direct figures from the subject brand but not audited in financial results.
How it works
Documented architectureThe stack in detail
- outil Sonantic realistic synthetic voice technology, acquired by Spotify in 2022, that gives the DJ its voice
- llm OpenAI generative AI technology cited by Spotify at the DJ launch to scale the commentary written by the newsroom
- infra Moteur de personnalisation Spotify in-house recommendation systems that choose which tracks to chain, reorientable by voice requests since 2025
How it runs, concretely
For ops teams-
1Music selection AI
The personalization engine chooses which tracks to chain for the listener, including any voice requests.
-
2Commentary writing editorial team / AI
Music experts write contextual scripts about artists and tracks; generative AI scales and personalizes them.
-
3Voice synthesis AI
The Sonantic voice turns the script into realistic audio, inserted between tracks.
-
4Listening and reorientation human / AI
The listener listens and can request another genre or mood out loud; the session recomposes.
The listener's listening history and preferences, which feed both the track selection and the choice of commentary. Without this signal, the DJ falls back to a generic radio with no personalization.
How your customers perceive this type of use
Sourced studiesUn ecart net separe les annonceurs des consommateurs : 77% des annonceurs voient l'IA positivement contre 38% des consommateurs (Yahoo/Publicis, 2024). Les mesures implicites confirment le rejet declare : en EEG, les pubs generees par IA produisent une activation memorielle plus faible que les pubs traditionnelles et sont decrites comme agacantes, ennuyeuses et confuses (NIQ, 2024). La disclosure a un effet ambivalent : elle augmente fortement la confiance quand elle est remarquee (Yahoo/Publicis), mais 27% des jeunes consommateurs disent faire moins confiance a une entreprise dont la pub est creee par IA (IAB, 2024).
Acceptance conditions
- Une disclosure visible : quand la mention IA est remarquee, la confiance globale envers l'entreprise augmente de 96% (Yahoo/Publicis 2024)
- Une qualite visuelle suffisante : les visuels IA de basse qualite augmentent l'effort cognitif et distraient du message (NIQ 2024)
Red lines
- Le contenu IA non declare puis identifie : 72% des consommateurs disent que l'IA rend l'authenticite difficile a etablir (Yahoo/Publicis 2024) et les marques utilisant des pubs IA sont plus souvent jugees inauthentiques ou non ethiques par les consommateurs que par les dirigeants (IAB 2024)
- Les mannequins et personnes generes par IA : 46% des consommateurs n'en veulent pas dans la publicite, l'inquietude premiere etant les standards de beaute irrealistes (Attest 2025)
Sources: Yahoo / Publicis Media (terrain Ebco) 2024 · IAB (avec Attest) 2024 · NIQ (NielsenIQ) 2024 · Attest 2025
How to replicate
Inference, not sourcedData prerequisites
- per-user listening / consumption history
- catalog enriched with editorial metadata
- corpus of scripts per artist / track
Org prerequisites
- editorial team for the scripts
- licensed voice synthesis capacity
- existing recommendation engine
Possible stack
- generative voice (Sonantic, ElevenLabs or equivalent)
- LLM to shape the commentary
- custom or managed recommendation engine
The plan, step by step
- Step 1Frame the editorial pipeline: who writes the scripts (experts, curators) and in what format per artist and trackDeliverable: Pilot script corpus and editorial guidelines
- Step 2Expose the track selection from the existing recommendation engine in a continuous session formatDeliverable: Playable personalized stream, without voice
- Step 3Scale the scripts with an LLM and synthesize the voice under license (Sonantic, ElevenLabs or equivalent)Deliverable: Voice DJ prototype on a limited catalog
- Step 4Closed beta on a panel: measure share of listening time and next-day returnDeliverable: Engagement metrics compared to the playlist baseline
- Step 5Roll out by markets and languages, then add user controls (voice requests)Deliverable: Feature in production with an engagement dashboard
First step: Frame the editorial pipeline (who writes the scripts) before the voice layer, because it is the quality of the commentary that makes the difference.
Sources
- S1 Behind the Scenes of Spotify's New AI DJ (Spotify Newsroom) Interested party archive pending
- S2 Spotify's DJ Now Takes Requests, Enhancing Real-Time Music Discovery (Spotify Newsroom) Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.