You are listening to a track and something feels off. The vocal is a little too even, the cymbals a little too clean. You search the artist and find nothing. So you ask the obvious question: is this song AI?
The platforms have started answering that question, but each one answers a different version of it. One labels the artist. One labels the track. Two wait for someone to declare it. If you read all four the same way, you will draw the wrong conclusion.
This page covers what each service actually shows a listener in November 2026, and then the three checks you can run yourself when no label appears.
We wrote the policy side separately in Do Spotify, Deezer or YouTube Flag AI Music? That page is about what the platforms do to AI music. This one is about what you can see.
The short answer
A label is a declaration, not a measurement. Spotify badges artist identities, Deezer labels tracks with its own detector, and YouTube and Apple Music show a label only when the uploader or the distributor supplied one. An unlabelled track is not evidence that a human made it.
A label is a declaration, not a measurement
There are two different things a platform can put on a song.
The first is a declaration: someone in the supply chain says how the track was made. The distributor fills in a metadata field, or the creator ticks a box at upload. The platform displays what it was told.
The second is a measurement: the platform analyses the audio and decides for itself.
Of the four services below, only Deezer says it does the second one at scale. The rest depend on the first. That difference decides what a missing label means, and it is the reason platform labels and audio detection answer different questions.
Spotify: the badge is on the artist, not the track
Spotify announced AI Persona badges on 11 August 2026. The badge sits on artist profiles, in search results, and on track rows in playlists. Tap it and Spotify shows whether the artist disclosed the persona themselves or whether Spotify's review team made the determination.
Two details matter more than the badge itself. Spotify said it would not rely on self-disclosure alone, and that by default it will not include AI Personas in editorial or algorithmic recommendations unless a listener follows them. The company also lists AI Credits among the tools it launched for artists in 2026.
What this tells you: whether the artist identity looks AI-generated. It does not tell you whether a particular track was generated, and a real artist using AI tools can carry no badge at all.
Deezer: the label is on the track
Deezer is the one service of the four that describes automated detection of the audio itself. Its own page says a "patent-pending" system scans for signatures left by generative platforms such as Suno and Udio and tags the track automatically. Flagged tracks are excluded from Flow and from personalised and editorial playlists.
Deezer also publishes scale: about 13 million songs tagged in 2025, and roughly 75,000 AI-generated tracks uploaded per day as of April 2026.
What this tells you: Deezer thinks this specific track was generated. It is a measurement, so it can be wrong in both directions. A false positive is a real track tagged as AI, and no detector on the market has a zero error rate.
YouTube: the label exists only if the uploader ticked it
YouTube has required creators to disclose realistic altered or synthetic content since March 2024. The disclosure produces a label, and YouTube's own help pages list AI-generated music as an example of content that needs one. Content that is clearly unrealistic or animated, or that used generative AI only as a production aid, does not require disclosure.
What this tells you: if the label is there, the uploader said so. If it is missing, nobody said anything, and that is all it means. Most AI music on YouTube is audio with a still image, which is exactly the case where a creator may reasonably decide no disclosure is needed.
Apple Music: the label comes from the distributor
Apple Music told its industry partners in August 2026 that it would add a "Made With AI" label to songs where "a material portion of the content" was created with AI, and that labels and distributors would be required to tag those songs. The label follows AI Transparency Tagging, which Apple introduced in March 2026. Variety reported the plan on 20 August 2026, including that the timing of the visible labels was not yet fixed.
What this tells you: the same as YouTube, one step further up the chain. A distributor decides, and Apple displays it. Apple Music executive Oliver Schusser told Billboard earlier in 2026 that more than a third of tracks uploaded to the service were "100% AI", which is a fair indication of why the metadata route exists.
Tidal, for completeness
Tidal has said it will label AI music and bar those tracks from accruing royalties. That combination is the strictest of the services here, and it is worth knowing that a label on one platform does not imply a label on another.
What none of these labels can tell you
- Whether this specific file is AI. Only Deezer measures the audio, and any detector has false positives and false negatives.
- Whether a human was involved. A track can be AI-generated and heavily edited by a person, or human-written and processed with AI tools.
- Whether the track is registered as AI elsewhere. Metadata does not travel between platforms by itself.
- Whether you are looking at the same master. A distributor can deliver one version and a video host can carry another.
Three checks you can run yourself
When no label appears, these are the checks that work on any platform, in order of cost.
1. Listen to the tail, not the attack. Generative vocals often match the onset of a note and then lose texture as it decays. Listen to the last 200 ms of a phrase rather than the first. This is a judgement, not a test, and trained listeners get it wrong too.
2. Look at the spectrum. A flat cut near the top of the frequency range usually means lossy encoding, not AI. It tells you about the codec, not about authorship. WAV vs MP3 for AI music covers what those differences look like.
3. Run a detector and read the score honestly. A score is evidence for review, not proof of authorship. Pair every score with the detector's false-positive rate, and check which file and which part of it was scored. Our free detector scores a track in segments and needs no account: demo.intrect.io. If you want to compare several detectors on one test set first, our 2026 comparison runs eight of them on the same 2,104 files.
If you need the theory behind the third check, How to Detect AI-Generated Music covers the methods, and provenance, watermarking and detection explains why they are three different things that people keep conflating.
What we make
We build ArtifactNet, the detector behind the free tool linked above, and audio-repair plug-ins that reduce artifacts in generated music. That is a commercial interest and you should weigh it. Nothing in this page depends on our numbers: the platform behaviour above is quoted from each company's own pages and from trade reporting, and every one of those sources is linked below.
How we keep this page current
Platform labelling is moving. We re-read the five sources below once a quarter and update this page when a mechanism changes, with the review date in the page metadata. The last check was 6 November 2026.
A score is evidence, not a verdict
Run any track through the free ArtifactNet demo and read the segment scores before you decide anything.
Related guides
- Do Spotify, Deezer or YouTube Flag AI Music?, the policy side
- How to Detect AI-Generated Music, the three methods
- Best AI Music Detector (2026), eight detectors on one test set
- Audio Provenance vs AI Detection vs Watermarking, why they differ
Sources
- Spotify newsroom, "Introducing a New Label for AI-Generated Artist Identities", 11 August 2026. newsroom.spotify.com
- Deezer, "Labeling AI-generated music", read 6 November 2026. deezer.com
- YouTube Blog, "How we're helping creators disclose altered or synthetic content", 18 March 2024. blog.youtube
- YouTube Help, "Disclosing use of GenAI content". support.google.com
- Variety, "Apple Music to Label AI-Made Tracks Later This Year", 20 August 2026. variety.com
FAQ
Does Spotify tell you if a song is AI?
Not per track. Spotify's AI Persona badge applies to artist identities and appears on profiles, in search and on track rows. It signals that an artist's identity may be AI-generated, not that a particular recording was.
Which platform actually detects AI music from the audio?
Deezer describes automated detection that scans for signatures from generative platforms and tags the track, and it excludes flagged tracks from Flow and its playlists. The other services described here rely on disclosure from creators, labels or distributors.
Is an unlabelled song proof that a human made it?
No. Two of the four services only show a label when someone declared AI use, so a missing label means no declaration was made. Use the audio checks above, and treat any single result as evidence rather than proof.
Can Apple Music's Made With AI label be wrong?
The label comes from metadata supplied by labels and distributors rather than from audio analysis, so its accuracy depends on what the supplier declares. Apple said in August 2026 that it would require suppliers to tag songs where a material portion of the content was created with AI.