Spotify has begun letting artists declare themselves as AI personas, introducing a label intended to help listeners distinguish human musicians from artificial ones. The scheme is voluntary, which is both its central feature and its main limitation.
Artists have been able to apply the AI Persona tag to their own profiles through the Spotify for Artists portal since 11 August. Listeners will only start seeing the labels from mid-September.
Spotify says it hopes artists will be transparent, but is not relying on that alone. The platform will run regular checks using a combination of human review and AI-powered detection to identify accounts that should carry the label but have not applied it. Artists flagged this way will be notified and can appeal if the determination was made in error. By default, accounts flagged as AI without self-declaring will not appear in editorial or algorithmic recommendations.
The profile display will also distinguish between the two routes — showing whether an account declared itself or was caught. For listeners deciding what to play, that distinction may matter more than the label itself.
What the label does not address
For African musicians and rights holders, the labelling scheme sits at an angle to the questions that have dominated the past year’s debate.
At the ICT Editors Xchange in Johannesburg in July, AuraaAfrica founder Gift Lubele told delegates that 94% of AI music models are trained on Western data, with African data accounting for 0.04% — not even a full percentage point. “If we don’t include ourselves in these AI models, we cease to exist,” he said.
Spotify’s label answers whether a track was generated by AI. It says nothing about what trained the model, whether the training data was licensed, or whether anyone whose work contributed to it was compensated. Those are the questions that surfaced repeatedly at the Wits AI and African Music showcase in June, where five artist-engineer teams from seven African countries presented work, and where panellists returned to the problem of who owns African musical material once it enters a training set.
Ninel Musson, co-founder of Music Business Lab and founder of independent label Vth Season, made the structural point at that showcase: the challenge is compounded by the fact that most distribution platforms are not African. A voluntary disclosure policy designed in Stockholm and applied worldwide is an illustration of exactly that dynamic — African artists and audiences are subject to rules they had no part in setting.
Where enforcement will be tested
The scheme’s effectiveness depends on detection, since bad-faith operators are unlikely to self-declare. Spotify has not published detail on how its detection systems work or how accurate they are.
That matters for African markets in a particular way. Detection systems trained predominantly on Western musical forms may perform less reliably on Afrobeats, amapiano, kwaito, highlife or the range of regional genres that dominate African streaming — the same underlying problem Lubele’s 0.04% figure describes, appearing on the enforcement side rather than the generation side. A system that cannot reliably identify AI-generated amapiano is a system that under-protects amapiano artists.
Ugandan producer Benon Mugumbya described the human-detection version of the same test last month, telling reporters that AI-generated music remains recognisable by being too clean — perfect vocals, crystalline production, no rough edges. “It’s still missing that soul,” he said. Whether Spotify’s automated systems can make that judgment across genres they have seen relatively little of is an open question.
For now, the label is a disclosure mechanism rather than a rights framework. It tells listeners something useful. It does not tell African artists whether their work is in the training data, and it does not pay them if it is.
