On Tuesday August 11, Spotify said it will start flagging artist profiles built around an AI-generated identity with an "AI Persona" badge, and keep the music behind those profiles out of personalized recommendations by default. Nothing has changed on the app yet. The badge will start appearing from mid-September. If you make podcasts or audiobooks, none of it binds you. It still tells you exactly where this is heading.
Here is the part worth being honest about first, because most coverage this month blurred it: every platform labelling move of the last few months has been a music policy. Not a podcast policy. Not an audiobook policy. The distinction matters, and so does the pattern the music side is establishing, because the two catalogues sit inside the same apps and get governed by the same teams.
What did Spotify actually announce on August 11?
Spotify is introducing a badge that will tell listeners an artist's identity "may be AI-generated and does not represent a real person", according to Music Business Worldwide's report of the August 11 announcement. From mid-September it will show up on mobile in four places: the profile banner, the About section, Search, and track rows across playlists.
Two mechanics will make it more than a cosmetic tag. First, music from AI Personas will not be included in any of Spotify's editorial or algorithmic recommendations by default, unless a listener chooses to follow the profile. Second, the badge is not purely self-declared. Artists can disclose through Spotify for Artists as of the announcement, and Spotify said it will also use human reviewers and what it calls "AI investigative tools" to apply it, starting with artists past "defined audience thresholds". Flagged artists will be notified and can then self-disclose or appeal.
Spotify was careful about scope: the badge concerns the public identity, the name and imagery on the profile, rather than how the music was made. For production disclosure it pointed to its AI Credits and SongDNA features.
Why this is a music policy and not a podcast policy
Read the announcement closely and the subject is an artist profile in a music catalogue. Podcasts are not mentioned. Audiobooks are not mentioned. Spotify has announced no equivalent badge for either.
That is worth stating plainly, because the reflex reaction to any Spotify AI headline in creator forums is "so my AI-narrated show gets buried now." No announced policy says that. The music side is a preview of the shape platforms are settling on, not a rule you are already breaking.
What else changed in music this July and August
The announcement was the visible end of a run of moves, all of them from the recording industry rather than the spoken side.
| Change | Who | When | What it does |
|---|---|---|---|
| Voluntary "AI-generated" and "AI-assisted" tags proposed | RIAA and IFPI | July 2026 | Two labels for AI-made recordings across services |
| Chart-eligibility principles | Three majors plus Believe, BMG, Concord, HYBE | July 29, 2026 | Bars AI-made recordings from official charts unless lawfully made and "substantially human-made" |
| Tag fully AI-generated music, block royalties | TIDAL | June 2026 | Identified tracks stop earning |
| AI transparency tags | Apple Music | March 2026 | Relies on labels and distributors to declare, not platform detection |
| AI Persona badge, deranked by default | Spotify | Announced August 11, 2026, appearing from mid-September | Will flag AI identities and keep them out of default recommendations |
IFPI has begun applying the chart principles across the official charts it runs and says it is expanding to more than 20 further charts, including Australia's ARIA Charts and Germany's Offizielle Deutsche Charts.
The volume context explains the urgency. Deezer says fully AI-generated tracks passed half of its daily uploads for the first time in June 2026, at close to 90,000 a day, up from 75,000 a day, or 44 percent, in April. Deezer also reports that up to 85 percent of streams on fully AI-generated music were fraudulent in 2025. Spotify said in September 2025 that it had removed more than 75 million "spammy" tracks over the prior year.
That last number is the tell. These policies are anti-fraud measures wearing transparency clothes. The problem platforms are solving is 90,000 daily uploads gaming royalty pools, not a novelist narrating her own book with a digital voice.
What the emerging pattern is: disclose, then get deranked
Strip the specifics and one shape repeats. A platform asks for disclosure, accepts self-declaration as the primary route, adds detection as a backstop, then applies a consequence to the undisclosed or fully synthetic case. The consequence is rarely removal. It is loss of distribution: no recommendations, no chart eligibility, no royalties.
That is a meaningful design choice. Platforms have decided AI content mostly does not need to be banned, it needs to be sorted. Which means the cost of not disclosing keeps climbing, while the cost of disclosing, at least in spoken audio today, is a sentence and one narrow distribution limit.
Does spoken audio already have AI labels?
It does, in a much gentler form, and it has for a while.
Spotify for Authors accepts audiobooks with digital voice narration. When you upload, you select the option marked "This audiobook uses digital voice narration", and Spotify adds a short sentence to the book's description so listeners know. Named accepted providers include Google Play Books auto-narration and ElevenLabs. There is one real limitation: Spotify does not currently share audiobooks with digital voice narration to referral partners. Replace the narration with a human voice and unselect the option, and the disclosure sentence comes off.
That referral-partner limit is a genuine cost, and it is the only one. No recommendation penalty, no royalty block, no badge on your profile, and a disclosure you can reverse by swapping the narration.
YouTube is even clearer. Its help page on disclosing GenAI content states, in a note under the labelling section: "Disclosing AI content won't limit a video's audience or impact its eligibility to earn money." The same page lists what needs no disclosure at all, and "Cloning one's own voice to create voice overs or dubs" is on that list, alongside voice or audio repair, upscaling, caption creation, and using AI for outlines, scripts, thumbnails and titles. What triggers disclosure there is realistic content about real people and real events, not synthetic narration as such.
Where the EU AI Act sits under all of it
None of this is happening in a vacuum. The EU AI Act's content-labelling rules took effect on August 2, 2026, and as Music Business Worldwide noted in its August 11 piece, those rules require whoever deploys an AI system to label deepfakes, a duty the European Commission says covers audio built to mimic a real artist's voice.
So the regulation sets a floor and the platforms are building above it. Spotify's badge will go further than the law requires for music identity. The Spotify for Authors sentence and YouTube's attributes field are the spoken-audio equivalents of the same instinct. Our walkthrough of what Article 50 actually asks creators to label covers the obligations in detail, and the YouTube-specific rules are worth reading before you touch the attributes checkbox.
What podcasters and audiobook producers should do now
Use the toggle where it exists. Spotify for Authors has one, YouTube Studio has one under Attributes. Selecting them costs you a sentence and buys you a documented good-faith position if policy tightens.
Keep provenance records. For each finished file, note which voice you used, whose consent you had if the voice is modelled on a person, and which script version produced the audio. This takes a spreadsheet. It is the single thing you will wish you had if a platform ever asks. Producing locally helps here, because the script, the voice and the render all stay on your machine where you can point at them. That is also the argument in our piece on the ethics of altered voices, which gets more practical every month.
Treat disclosure as table stakes. The music policies punish concealment and fraud far more than they punish synthesis. Nothing in the spoken-audio rules today penalises a disclosed digital narration. Write one plain line for your show notes, reuse it, stop worrying about it. Something like: this episode uses a digital voice for narration.
If you want the disclosure lines, consent language and voice notes already assembled, the podcast voice kit has them in a form you can paste. And when policy shifts again, which it will before the year is out, we send the readable version to the newsletter rather than making you read law-firm blogs.
The creators who will have an easy time of the next twelve months are the ones who already treat "yes, this is a digital voice" as a boring fact about their production. Say it early and it never becomes a story.
The Vois Team