The Tidal AI policy only touches music that is wholly generated by artificial intelligence, which Tidal defines as a track created entirely by AI with no human writing, performing or producing anywhere behind it. If a person wrote it, sang it or produced it, your release does not meet that definition. An AI mastering plugin does not meet it either. Tracks Tidal identifies as wholly AI generated get an AI badge shown to listeners, earn no royalties at all, and cannot be sold through direct to fan sales. Tidal set this out in June 2026 and the loss of royalties applied from that moment.
Table of Contents
The line Tidal drew is about whether a human was involved at all
Read the definition slowly, because one word carries the whole thing. Entirely. Machine learning has been living inside mastering chains, noise removal, pitch correction and stem separation for years, and none of that suddenly became a policy problem in June. What Tidal is describing is a track where no person wrote, played, sang or produced any part of it. A prompt goes in, a finished song comes out, and nobody touched the middle.
That distinction has a name worth learning, because you will need it when you talk to your distributor. AI assisted is a different category from AI generated. Almost every release made this decade is AI assisted somewhere. Very few are AI generated.
Tidal did add one thing that should keep you paying attention. It said it plans to extend the labelling to substantially AI generated music as its detection improves. Today the test is close to all or nothing. It will not stay that way permanently, so knowing where your own workflow sits is useful even if you are comfortably clear of the line right now.
Where the common workflows actually sit today
| What you did | Where it lands |
|---|---|
| AI mastering, an AI limiter, or AI noise removal on a song you played | Well outside the definition, because a human performed and produced it |
| AI stem separation or pitch correction used during the mix | Outside the definition, since these are processing tools rather than writers |
| A generated instrumental bed that you wrote and sang a topline over | Outside the definition, because human writing and performance are both present |
| Lyrics drafted with an AI assistant, then performed and produced by you | Outside the definition, since a human performance carries the recording |
| An AI voice model singing a song you wrote, with no human performance on the record | The uncomfortable middle, and the place the definition is likely to widen into |
| Prompt in, finished track out, delivered exactly as the generator produced it | Inside the definition |
If your track sits in the middle row, the honest thing to do is record something real onto it. A live vocal, a played bassline, anything performed by a person. That changes what the recording is rather than how you describe it.
What a tag actually costs you
Here is where most artists brace for the worst. The track gets pulled, the release disappears, the distributor drops you and years of catalogue go with it. You were expecting me to confirm all of that, and it is wrong. A wholly AI generated release is not removed from Tidal. It stays exactly where it is, with a badge beside it, earning nothing. No royalties are paid on it, and it cannot be sold direct to fans either.
Which is arguably worse than a takedown. A takedown is loud, and at least you know where you stand. A demonetised release looks entirely normal from the outside while every single play is worth zero, and nothing on your artist page tells you that has happened. Tidal Upload applies the same standard to anything you deliver yourself, and blocks uploads it suspects are wholly AI generated from direct to fan monetisation.
So the cost is money and it is also reach, because a badge changes how a listener reads your release before they have heard a second of it.
Your distributor is now the one making the declaration
The AI information reaches Tidal inside the delivery your distributor sends, sitting alongside the ISRC and your credits. Tidal has said it expects content distributors to identify AI generated content before it arrives on the platform, and that it will begin enforcing that expectation. So whatever field you tick at DistroKid, TuneCore, Ditto, CD Baby or wherever you release from is the thing Tidal reads first.
Two practical consequences follow. Find that field before your next delivery and answer it truthfully, because a declaration that contradicts what Tidal’s own systems conclude leaves you in a far worse position than an honest one ever could. And if your distributor has no AI field anywhere in its upload flow, write and ask them where the declaration goes, because their delivery is speaking on your behalf whether or not you were consulted.
Tidal is not relying on that declaration alone. It identifies content itself as well, which is the entire difference between this policy and an honour system.
Impersonation is a separate and much harsher track
Everything above is about labelling. There is a second category that gets content removed instead, and it deserves plain language.
Tidal removes AI generated music that exploits an individual’s or a group’s music, name or likeness, that deceives listeners, or that damages the quality of its service. In practice that means a voice model trained on a named artist and released as a song. It means a record built so that listeners mistake it for somebody else’s work. It means metadata engineered to land your upload on another artist’s profile page.
Those are removals, and they rarely stop at the release. Complaints of that kind usually arrive through your distributor, which means your whole distribution account is exposed, along with every other track you have out. If any part of your release depends on a listener believing you are somebody you are not, stop now and change it. This is the one place in the whole policy where there is no grey area to argue about later.
Tidal, Deezer and Apple Music are not asking you the same question
Tidal combines its own detection with a requirement that distributors declare. Deezer runs detection at the platform level, tags albums that contain fully AI generated tracks, and has been keeping fully AI generated content out of its algorithmic recommendations and its editorial playlists since early 2025. Apple leans on labels and distributors to declare it in the metadata they deliver.
The useful conclusion from those differences is that a single honest declaration, made once at your distributor, satisfies every one of them at the same time. A dishonest one has to survive the strictest detector on any platform you appear on, and Deezer has said its detector recognises output from the best known generators. Deezer has also reported that a large share of daily uploads are now fully AI generated and that a striking proportion of the plays on those tracks look fraudulent, which explains why the enforcement is tightening everywhere at once rather than only at Tidal.
What to do before your next release
- Write down every tool that touched the record and the stage it touched it at. You want this on paper before somebody asks.
- Keep your session files, stems and any recording of a real performance. That is your evidence of human involvement if a decision ever needs contesting.
- Fill in your distributor’s AI field honestly, and screenshot the confirmation.
- If a generated bed is the foundation of the track, perform something over it and keep the raw take.
- Never clone a real person’s voice for a commercial release, and never build metadata that leans on another artist’s name.
The short version for anyone still nervous. If a human wrote it, played it or sang it, the Tidal policy was not written about you. If nobody did, it was, and the consequence is a badge and an empty royalty line rather than a removal. Everything in between is a question about how you fill in one field at your distributor, which is a much smaller problem than the panic going around suggests.