TL;DR

Spotify is introducing an “AI persona” tag for artist identities generated by AI, displayed on profiles and in playlist track listings from September. Tagged acts will also be excluded by default from personalised recommendations. The platform removed 75 million spam tracks across a twelve-month period last year, against a legitimate catalogue of 100 million.

Demotion is the part that matters

Labelling on its own changes little — most listeners never inspect an artist profile. Withholding the recommendation slot is the substantive measure, because on a streaming service discovery is the distribution channel. An act excluded from personalised suggestions is not being censored; it is being made to find an audience without the algorithm’s help.

That is aimed squarely at displacement. Where synthetic acts absorb streams that would otherwise reach human musicians, the payout follows the play, and the platform is now putting a thumb on the scale.

Enforcement does not rest on the honour system. Creators may declare an identity as AI-built and receive the tag, but Spotify says it will also apply human review and its own investigative tooling to catch those who do not. The company treats an AI-generated profile as reliable evidence that the music behind it is machine-made too.

The cases that forced the issue

The Velvet Sundown, a group styled on 1970s American rock, spent months provoking argument about its origins before its own social accounts conceded the deception, describing the project as sitting “somewhere in between” human and machine. It draws 117,000 monthly listeners; its page still carries no explicit AI marker, describing the four-piece instead as a synthetic music project.

A second case followed in January, when the soul singer Sienna Rose reached the platform’s viral top 50 having never performed live or released a video, despite an unusually prolific ten-week run of releases. That profile — now past half a million monthly listeners — remains unclarified.

Looking forward

For British musicians and labels this lands in the middle of the copyright and transparency fight, and it establishes a workable precedent: disclosure enforced by the platform rather than requested from the uploader.

It also rhymes with what model developers have started doing at the other end of the pipeline. Anthropic confirmed this week that it will watermark Claude’s text output worldwide to satisfy European transparency requirements. Both moves accept the same premise — that provenance has to be machine-checkable, because asking people to self-certify has already failed.