Spotify built a badge, and quietly switched off the recommendations behind it

What Spotify actually shipped on 11 August#

The fox does not bark at a fence. It walks the length of it first, finds the gap, and decides whether the field beyond is worth the crossing. That patient prowl is the posture the music business needs this month, because the fence Spotify put up on 11 August is not where most people are looking.

Spotify opened self-disclosure for AI-generated artist identities and announced a badge, the AI Persona, that begins appearing on profiles from mid-September. Artists declare themselves through Spotify for Artists, and Spotify may also apply a Likely AI Persona badge itself after review, with an appeals route for anyone marked in error. All of that is presentation, and presentation is the part everybody read.

The sentence that actually moves money, and that rewrites music promotion for a whole class of act, sits further down the same announcement. Spotify states plainly that By default, Spotify will not include AI Personas in any editorial or algorithmic recommendations. Nothing is removed. Nothing is banned. The catalogue stays, and the distribution goes.

Disclosure is the trigger, distribution is the penalty#

The Spotify for Artists documentation repeats it in the operator's register, confirming that music from AI Personas will not be included in any editorial or algorithmic recommendations. It also notes that a self-applied badge cannot be removed by the artist, which makes the disclosure decision a one-way door rather than a toggle.

There is a second, quieter cost. Spotify's Verified by Spotify guidance says that profiles that appear to primarily represent AI-generated or AI-persona artists are not eligible for review or approval. So the badge does not merely mute the algorithm, it also locks the act out of the trust mark that sits above it. Two doors, one decision.

Why the badge is a discovery tax, not a takedown#

Removal is loud, litigable and rare. Demotion is silent, cheap and permanent, and it is the tool platforms reach for when they want a behaviour changed without a fight. Spotify has not banned synthetic acts. It has made them pay rent on a road that used to be free.

Understand what is being withdrawn. Editorial playlists, algorithmic playlists and personal mixes are the three engines that turn a release into a career, and they are the engines an AI Persona is switched out of by default. What remains is search, direct links, and whatever a marketing budget can buy. That is the whole shape of the change, and it is why music promotion for a labelled act stops being amplification and starts being acquisition.

This is the part agencies will get wrong. Most music promotion services are priced on the assumption that paid media buys a first push and the algorithm does the compounding afterwards. Strip out the compounding and the same budget delivers a fraction of the same result, because you are now renting every single listener rather than seeding a system that fetches them for you.

The honest arithmetic nobody wants to run#

Run the sums before the badge lands rather than after. If a campaign's modelled outcome assumes algorithmic pickup after week two, that assumption is now a line item with a price on it, and the price is whatever paid social and search cost in your market. Music promotion companies quoting on last year's model are quoting on a road that has been closed.

The fox reads a hedgerow by where the gaps are, not by how tall it stands. The gap here is search and direct traffic, both of which survive the badge untouched, and both of which reward exactly the owned-asset work that gets deprioritised when playlists are doing the heavy lifting.

The flood this badge is answering#

None of this is a moral panic invented by a platform team. The supply side moved first, and it moved fast enough that a labelling regime became inevitable rather than optional.

Deezer, which has published its detection figures more openly than any other service, reported in July 2026 that 90,000 AI-generated tracks per day now represent over 50% of all new music uploads. Eight months earlier the same newsroom put the daily figure at roughly 50,000 fully AI-generated tracks are now uploaded every day. That is the curve a badge is built to bend.

The listener side of the same research matters just as much to anyone selling music promotion. Deezer's survey with Ipsos, across 9,000 respondents in eight countries, found that most people cannot tell the difference unaided, which is precisely why a visible label changes behaviour where an invisible watermark does not.

The tools are labelling themselves too#

Generators are moving in the same direction, and on a nearer deadline. Suno told users that On September 3, we will introduce the following download limits based on your tier, capping exports at 7 lifetime downloads on the free tier, 20 a month on Pro and 60 a month on Premier. In a companion post it confirmed that we will also be adopting new audio watermarking and fingerprinting technology.

Udio went further still after its Universal settlement, telling users that downloading of audio, video, and stems has been disabled. Suno metered the door. Udio closed it. Either way, the era of unlimited free synthetic assets for a campaign has quietly ended.

There is law arriving behind the platforms as well. The EU AI Act's Article 50 requires synthetic output to be marked in a machine-readable format and detectable as artificially generated or manipulated, which is why watermarking suddenly appeared on generator roadmaps this summer rather than next year.

And the rights holders are not waiting either. Germany's collecting society GEMA, suing over training data, put its position bluntly: Anyone who wants to use these songs must acquire a license and remunerate the authors fairly. A CISAC study using case studies and expert interviews puts 24% and 21% of their revenues at risk of loss by 2028 for music and audiovisual creators under an unchanged-regulation scenario.

Five moves for music promotion after the badge#

Here is the practical part, and none of it requires knowing whether your act will be badged. Each move makes music promotion sturdier whichever way the badge falls. Every move below is worth making either way, which is the test folkfox applies to any recommendation made under uncertainty.

Do not skip the boring one#

The fifth move is the one that gets postponed, because documentation feels like admin rather than marketing. It is the only move on the list that is evidence. A badge applied by review is contested with records, not with indignation, and the appeals process Spotify describes will be won by whoever can show their working.

Apple's approach is narrower and more technical, a delivery-spec field for disclosing when artificial intelligence is used to generate a material portion of a music video single. YouTube's is a viewer-facing label where For most videos, a label will appear in the expanded description. Neither attaches a distribution penalty. Spotify's does, and that is the whole story.

Underneath all of them sits DDEX, the industry's delivery plumbing, which now carries a signal describing the amount of AI that has been involved in the creation of a particular sound recording or music video. Once that field is populated at distribution, every downstream service can act on it without asking the artist again.

Measuring music promotion when the algorithm stops helping#

Reporting built on playlist adds is reporting built on a mechanism that may no longer apply to your act, which makes most music promotion dashboards quietly obsolete. The measurement has to move to the surfaces the badge does not touch, and it has to move before the badge arrives, or you will have no baseline to compare against.

Take the snapshot now. Capture monthly listeners, save rate, source-of-stream split and search-driven traffic for every act in the roster this week, while recommendations are still flowing normally. A change you cannot show is a change no label will fund a response to. The fox does not guess the depth of a stream, it puts a paw in first.

Treat detection accuracy with the care its own authors do. The Deezer Research team behind those numbers reported achieving a convincing accuracy of 99.8% while stating plainly in the same paper that calibration and robustness remain open problems. A near-perfect score in a controlled evaluation is not a promise about your act's appeal, and the appeals process exists precisely because classifiers are fallible.

One number to take to a label meeting: the share of monthly listeners arriving from surfaces the badge does not control. Search, direct and paid. If that share is under half, the roster is exposed, and the exposure is quantifiable rather than vibes. That single line turns an anxious conversation about AI into a budget conversation about channels, which is the only version of it anybody can act on.

Trade press read the announcement the same way. Music Ally noted simply that Spotify says there will be an appeals process. True, and useful, and beside the point for a marketer: the appeal restores a badge decision, never the eight weeks of algorithmic momentum lost while it was pending.

If you would rather have the measurement layer built than described, that is what folkfox music industry marketing does, alongside the content marketing and SEO and GEO work that keeps search and owned surfaces earning while the recommendation engine is switched off. Where paid has to carry the whole release, that is paid social and paid search doing work the playlists used to do free. The same disclosure logic is already reshaping brand strategy across every regulated category folkfox works in.

Frequently asked questions#