Skip to content
FREN
Menu

Platform rules

Platform rules

How streaming platforms and distributors detect, label, block or demonetise AI-generated music, and what options artists have for appealing their decisions.

Over eighteen months, AI-generated music has gone from having no rules at all to a patchwork of policies, with each service drawing its own line. Some services rely on a detector, others on disclosures by record labels and distributors, and a few on both. Depending on the service, a track identified as AI-generated may be labelled, lose its revenue or be removed from the catalogue. Yet almost no platform publishes an appeal procedure worthy of the name, with a clear point of contact, details of the evidence accepted and a timeframe for a response, and almost none has had its accuracy figures checked by anyone outside the company.

This page sets out what each platform and distributor says it does. Where a rule is undocumented, we say so rather than speculate, and figures supplied by platforms are presented as claims rather than established facts. Most of these rules offer the least guidance in the grey area of hybrid creation, where an otherwise human-made track involves AI in a stem (a separate audio track), in mastering or in developing an idea.

At a glance

Platform AI detection Labelling Blocking or demonetisation Appeal route Reported accuracy Last known change
Spotify Partly, through an anti-spam filter and reviews of "AI Persona" profiles "AI Persona" badge on profiles; AI use disclosed in track credits through the DDEX metadata standard Music from AI Persona profiles excluded from recommendations; unauthorised voice clones removed; no demonetisation solely for using AI Appeals through Spotify for Artists; badge criteria published, but no published list of accepted evidence or response timeframe None published 23/09/2026 (AI Persona help page)
Apple Music Not used for labelling; internal tools mentioned but undocumented "AI Transparency Tags" supplied at delivery; "Made With AI" label announced for listeners No documented demonetisation; exclusion from editorial playlists reported but unverified Not documented None 08/2026 (tags set to become mandatory)
Deezer Yes, with an in-house detector AI label applied at album level Excluded from algorithmic and editorial recommendations; fraudulent streams demonetised Through the distributor, then a Deezer support ticket if no response within 5 working days; manual review 99.8%, fewer than 1 false positive in 10,000 (figures supplied by Deezer, no independent audit published) 21/07/2026
TIDAL Yes, with automated tools "AI" badge on fully AI-generated tracks No royalties (since 29/06/2026) or direct-to-fan sales for fully AI-generated tracks; releases that impersonate artists removed TIDAL support form, with no published list of accepted evidence or response timeframe None; TIDAL acknowledges the limits of its detection 31/08/2026 (AI policy updated)
Qobuz Yes, with an in-house tool AI label applied at release level Excluded from editorial and algorithmic recommendations; AI-generated albums with no streams removed from search; fraudulent streams unpaid Through the distributor None; Qobuz acknowledges there is no reliable solution for music generated partly by AI 24/09/2026
Amazon Music Not documented None Takedowns on intellectual property grounds reported; no published AI policy Not documented None No published rules
YouTube Music Yes, with its own tools, alongside declarations from its music partners Label on the player page for tracks where generative AI played a significant role (composition, vocals); "altered or synthetic content" label on videos AI use alone does not make a track ineligible for monetisation; mass-produced AI content ineligible (video) Automatic labels can be challenged in YouTube Studio (video); no route announced for track labels None 14/10/2026 (track labels)
SoundCloud Not documented None; disclosure in the track description encouraged AI-generated tracks accepted for distribution through SoundCloud for Artists provided the relevant rights can be verified; impersonation through voice cloning prohibited SoundCloud for Artists support, with no published response timeframe None 28/07/2026 (AI disclosure encouraged)
Bandcamp No automated tools documented by Bandcamp, which relies on human review of fan reports; press reports say it uses the SH Labs detector None Music generated "wholly or in substantial part" by AI removed Not documented None 13/01/2026
Traxsource Yes, through SH Labs and SoundPatrol "Human-Made" or "AI-Assisted" Fully AI-generated tracks removed Appeal procedure for content providers; details unpublished None 01/07/2026
Beatport Yes, through Beatdapp, which distinguishes fully generated from AI-assisted tracks AI-assisted tracks accepted and tagged at ingestion for the curation team; no public label announced Fully or majority AI-generated tracks withheld at ingestion, with rightsholders notified Not documented Not documented 12/08/2026
DistroKid Conflicting reports; nothing documented by DistroKid AI use disclosed at upload and passed on to Spotify, Apple Music and YouTube Accepts AI-generated music if the uploader holds the rights; a secondary source citing no evidence describes review followed by removal of undeclared AI-generated tracks; nothing documented by DistroKid Not documented None 24/09/2026 (help page on AI credits)
CD Baby Yes according to secondary sources; nothing documented by CD Baby None Rejects content generated wholly or partly by AI; accounts excluded after repeated violations None documented None 2026 (undated help page)
TuneCore (Believe) Yes, with Believe's technology None Blocks tracks made with unlicensed generators (Suno models predating the deal); accepts AI-assisted music and music made with licensed generators Not documented "99% reliable" at identifying the model used (chief executive's statement, no independent audit published) 08/09/2026 (Suno deal)
Believe Yes, with in-house technology None Same as TuneCore Not documented "99% reliable" (chief executive's statement, no independent audit published) 08/09/2026
iMusician Case-by-case quality review None Fully AI-generated content rejected, with no refund Helps artists appeal to streaming platforms (DSPs), but cannot override their decisions None 11/08/2026
LabelGrid Pre-distribution review that checks the AI disclosure; methods not published Mandatory disclosure in three fields (artwork, sound recording, composition) on a four-level scale (none, minor, material, entirely AI-generated), sent to streaming platforms Entirely AI-generated tracks not delivered to YouTube Content ID, YouTube Music, Meta, TikTok or SoundCloud monetisation, with the rest of the distribution going ahead; AI voices imitating real artists not allowed Release sent back: fixed and resubmitted; release rejected: appeal to support with the release ID, the rejection reason and proof of rights, no published response time None 21/10/2026 (help pages)
Ditto Music For publishing only, where fully AI-generated tracks are identified by "our system"; not documented for distribution Not documented AI music accepted for distribution if the rights are held, with no voice cloning or bulk uploads; fully AI-generated tracks excluded from the publishing service Not documented Not documented 04/08/2026 (distribution help page)
IDOL Not documented Not documented No published rule on distributing AI-generated tracks; catalogue rights reserved against unlicensed AI training Not documented Not documented 22/11/2024 (AI statement, dated only in the page's metadata)
ONErpm Not documented Mandatory declaration at upload, on four levels (no AI, AI-assisted, partially AI-generated, fully AI-generated) None: all tracks remain eligible for monetisation, the declaration being "informational only" Not documented Not documented 28/02/2026
RouteNote Not documented; releases containing AI may undergo additional review Not documented AI music accepted under conditions; AI-generated content not delivered to content recognition services, Korean stores or Amazon Not documented Not documented 17/08/2026
Wiseband Not documented Not documented Reserves the right to reject AI-generated content "at our discretion" None: Wiseband says it is "pointless and futile" to contact it about rejections Not documented 24/09/2026 (update to an article of 03/01/2024)

Streaming platforms

Spotify

Spotify does not impose a blanket ban on AI-generated music and says tracks are not ranked lower solely because AI was used to make them. Since September 2025, its approach has combined an impersonation policy allowing vocal imitation only with the imitated artist's permission, an anti-spam filter targeting mass uploads and metadata manipulation, and support for the DDEX standard for disclosing AI use. With the beta launch of "AI Credits" on 16/04/2026, with DistroKid as the first distribution partner, a disclosure appears in a track's credits on mobile whenever a distributor reports that AI was used.

On 11/08/2026, Spotify announced an "AI Persona" badge for artist profiles. In use since mid-September 2026, the badge is assigned on the basis of both self-disclosure and Spotify's review of names and imagery, and music from profiles carrying it is excluded by default from editorial, algorithmic and personalised recommendations. Spotify says artists can appeal through Spotify for Artists. Its help page, updated on 23/09/2026, now sets out the badge criteria, which apply to profiles whose AI-generated imagery depicts a nonexistent person photorealistically. For appeals, however, the page refers artists to Spotify for Artists without specifying what evidence to provide or when to expect a response, and states that Spotify makes "all final decisions". No accuracy figures have been released for either the anti-spam filter or the profile reviews.

Sources:

Apple Music

Apple Music relies on disclosure for labelling. Introduced in March 2026, its "AI Transparency Tags" cover four categories (artwork, track, composition, music video) and are supplied by record labels and distributors when they deliver content, "similar to genres, credits and other metadata". Apple's specification states that no tag is assumed if none is supplied. On 20/08/2026, Apple told its content providers that the tags would become mandatory later in 2026 whenever AI had created a "material portion" of the content. It also announced a "Made With AI" label for listeners, due later in 2026.

Apple has not defined a threshold for that "material portion" or described any mechanism for checking compliance, and no appeal procedure is documented, since Apple does not assign the tags itself. The company has mentioned internal tools for identifying AI-generated tracks and manipulation, without documenting their role, scope or accuracy. Reports that AI-tagged content is excluded from editorial playlists remain unverified.

Sources:

Deezer

Of all the platforms, Deezer has the best-documented system. Its in-house detector examines every delivery track by track and is designed to identify fully AI-generated tracks, excluding AI used as a production tool from its scope. The visible label is applied to the album, a choice Deezer explains in these terms: "to reduce errors further, the AI label is applied at album level rather than at track level" (our translation, Deezer newsroom, 21/07/2026). Labelled tracks remain available but are excluded from algorithmic and editorial recommendations, and streams judged fraudulent generate no payment.

On 21/07/2026, Deezer announced that at the June 2026 peak, fully AI-generated tracks had accounted for more than 50% of daily deliveries, around 90,000 tracks a day, while making up only 1 to 3% of streams. The platform claims 99.8% accuracy, with fewer than 1 false positive per 10,000 human-made tracks. These figures come from Deezer itself, which reports tests with partners such as Sacem. No independent audit of the figures has been published.

To challenge a label, an artist first contacts their distributor, which passes the case to Deezer. If no response arrives within 5 working days, the artist can open a support ticket and attach their correspondence with the distributor. Deezer then promises a manual review, but publishes neither a list of accepted evidence nor a response timeframe.

Sources:

TIDAL

TIDAL published its AI policy on 29/06/2026 and stopped paying royalties that day on tracks it identifies as fully AI-generated. These tracks have also carried an "AI" badge since mid-July. They remain available but cannot be sold directly to fans, while releases that impersonate real artists are removed. AI-assisted music is explicitly allowed as long as it includes a human contribution. Yet the boundary remains unclear. TIDAL has not explained how it distinguishes fully AI-generated music from hybrid creation, such as a track that uses AI for a single stem or vocal part. The company acknowledges the limits of its detection, which is why the policy currently targets only fully AI-generated tracks, and intends to extend it "as detection improves". Since July 2026, its help page has offered artists a way to challenge a badge through the support form, but TIDAL has published no accuracy figures, list of accepted evidence or response timeframe.

Sources:

Qobuz

Qobuz published an AI transparency charter on 03/02/2026. Since late September 2026, it has displayed an AI label at release level on albums identified as AI-generated by its in-house tool. These albums are excluded from editorial and algorithmic recommendations; more than a third of the identified AI-generated albums with no streams have also been removed from search. According to Qobuz, 60% of streams of tracks identified as AI-generated are judged fraudulent and receive no payment. In version 2 of its charter, published in September 2026, the company openly acknowledges that "no reliable solution exists today to identify all AI content with certainty, particularly where partial human intervention has taken place" (our translation). Artists who challenge a badge must go through their distributor, which handles the request. Qobuz publishes neither accuracy figures nor a response timeframe.

Sources:

Amazon Music

Amazon Music has published no policy on AI-generated music and introduced no AI label. It has not documented any use of AI detection. The platform nonetheless hosts AI-generated tracks, has partnered with record labels to combat unlawful voice cloning, and has said it is "part of the conversations" about the labelling framework announced by the industry in July 2026. Artists have reported takedowns on intellectual property grounds, but these accounts come from secondary sources and remain unverified. With no documented appeal procedure, contact details or list of accepted evidence, artists whose tracks are removed have only their distributors to turn to. Amazon's position on AI-generated music is harder to understand given its announcement on 26/02/2025, at the launch of Alexa+, that it would integrate the Suno music generator into its assistant.

Sources:

YouTube Music

For videos, YouTube Music follows YouTube's policies, which require creators to apply the "altered or synthetic content" label to videos, particularly when AI generates the music central to the video or makes a real person appear to say or sing something they never said or sang (YouTube Help). Since May 2026, the label has appeared directly beneath the player, and YouTube also applies it automatically when its systems detect substantial use of photorealistic AI-generated content. Creators can challenge an automatic label in YouTube Studio unless C2PA metadata, which follows a technical standard for content provenance, identifies the content as fully AI-generated. Repeated failures to disclose AI use can lead to removal or suspension from the YouTube Partner Program. Since the programme's July 2025 update on repetitive and mass-produced content, AI content produced in bulk without human input has been ineligible for monetisation.

For the music itself, YouTube announced on 14/10/2026 a label shown on the YouTube Music player page when AI played a significant role in a track, for example when a generative model created its composition and vocals, or when the track was made primarily with generative AI. Tapping the label opens more details, and the feature is due to extend to the main YouTube app. YouTube says it relies both on its music partners and on its own detection tools. According to TechCrunch, partners declare AI use by choosing one of three designations (fully generated, partly generated, or no generative AI, the last covering edits such as Auto-Tune), and AI involvement alone does not make a track ineligible for monetisation. Neither YouTube's post nor the article describes a way to contest a label applied by detection, and no accuracy figures have been published.

Sources:

SoundCloud

AI-generated content can be uploaded to a SoundCloud profile without a ban, a label or any documented detection. Since 28/07/2026, however, SoundCloud has encouraged creators subject to the EU AI Act to state in the track description whether it was generated or manipulated by AI. It also reminds them that using voice-cloning tools to imitate an artist without permission remains prohibited. For distribution and monetisation, SoundCloud for Artists says it accepts AI-generated tracks provided the relevant rights can be verified, using the AI tool's terms of use if necessary. An earlier instruction that excluded content made with Suno, according to secondary sources, is no longer publicly accessible, and the current page on distribution rejections names no generator. SoundCloud's May 2025 terms of use update addresses model training. It commits the company to obtaining creators' explicit opt-in consent before using uploaded content to train generative models that would reproduce their voice, music or likeness. If distribution is refused, artists can contact support and attach their licences, although no response timeframe is published. There is no detection accuracy claim to assess, since SoundCloud makes no claim to detect AI-generated content.

Sources:

Bandcamp

With "Keeping Bandcamp Human", published on 13/01/2026, Bandcamp closed its doors to music and audio generated "wholly or in substantial part" by AI. It also placed the use of AI to imitate other artists under its rules against impersonation. People enforce these rules, with fans encouraged to report content for review by a dedicated team, and Bandcamp documents no detector. Yet Mixmag Asia reported on 22/08/2026 that SubmitHub's SH Labs detector is used by Bandcamp, as well as by Traxsource and the German collecting society GEMA. To our knowledge, Bandcamp says nothing about this on its own pages, so we do not know whether the tool plays a part in its decisions, or at what stage. With no definition of "substantial part", the treatment of hybrid creation remains unclear. Bandcamp has said it will adapt the policy as the landscape changes, but no appeal or reinstatement procedure is documented.

Sources:

Traxsource

Traxsource is the only store that explicitly labels hybrid creation. Since 01/07/2026, every track has carried either a "Human-Made" or an "AI-Assisted" label, while fully AI-generated music is flagged for removal. Two detection partners, SH Labs and SoundPatrol, assign the classifications without a declaration from the people submitting the music. A track's label therefore reflects a detection provider's assessment rather than its producer's account. Traxsource says it offers an appeal procedure for content providers who believe their music has been mislabelled, but publishes no form, contact details, evidence requirements or response timeframe. No accuracy figures have been released for either partner, and neither has been independently audited.

Sources:

Beatport

On 12/08/2026, Beatport announced that it was expanding its partnership with Beatdapp, a music fraud detection specialist it had already been working with on streaming fraud, to identify AI-generated music across its catalogue. Under the store's updated content guidelines, fully or majority AI-generated tracks are no longer permitted and are withheld during ingestion, with rightsholders notified directly. Music created with AI assistance is still accepted, provided the finished track remains "majority human-made", and such releases are tagged at ingestion so that Beatport's curation team knows what it is reviewing and featuring. The announcement makes no mention of a label visible to buyers. Beatport says Beatdapp's technology can tell a fully generated track from an AI-assisted one, but gives no accuracy figures and does not explain how the "majority" share is measured. Nor does it say what a rightsholder can do if they dispute a decision to withhold a track.

Sources:

Distributors

DistroKid

DistroKid accepts music made with AI provided the uploader holds the rights, and prohibits impersonation and mass-generated spam. Since 16/04/2026, its upload process has included an AI disclosure step, which now distinguishes between lyrics, music and all or part of the audio. Initially introduced to supply Spotify's AI Credits through DDEX metadata, these credits are now passed on to Spotify, Apple Music and YouTube, according to DistroKid's help page updated on 24/09/2026. They can also be added or corrected after upload. Whether DistroKid uses its own detection tools remains unclear. One secondary source, which cites no evidence for its account, describes an automated scan that would hold undeclared tracks for review and remove those confirmed as AI-generated. Digital Music News, however, reports that the system relies on self-disclosure. DistroKid has documented neither account; its help page warns only that platforms may reject or remove releases that breach their rules, and it has published no accuracy figures. With no documented appeal procedure, the practical advice in circulation amounts to disclosing AI use when uploading.

Sources:

CD Baby

CD Baby has the most restrictive published policy. Its help centre says it cannot accept AI-generated content for distribution, even under a commercial licence and even when AI generated only part of the recording. The only stated exception is for AI production tools (mastering, mixing, stem separation) applied to an entirely human-made recording. Content judged to be AI-generated is withdrawn from distribution, and repeated attempts to submit it can lead CD Baby to end its relationship with the uploader. Secondary sources describe strict screening with a low threshold for flagging content, while CD Baby itself documents no detection method, accuracy figures or appeal procedure. Its help page is undated, so the date on which the policy took effect is also undocumented.

Sources:

TuneCore and Believe

Believe and TuneCore, its division for independent artists, block tracks made with generators they consider unlicensed, particularly Suno, which they described as a "pirate studio". The block was already in place when they announced it on 30/04/2026. They accept music made with licensed generators (ElevenLabs, Udio) and AI-assisted works that include a human contribution. At upload, TuneCore requires five declarations, including identification of the tools used. Believe's chief executive says its detection system is "99% reliable" at identifying the model used to generate a track, a claim that has not been audited.

Neither company has published an appeal procedure. The only advice in circulation, from Undetectr, suggests resubmitting the track under a new ISRC (International Standard Recording Code), which restarts the submission process without challenging the original decision. On 08/09/2026, Believe signed a licensing deal with Suno limited to Suno's new partner model, so tracks made with earlier models remain blocked.

Sources:

iMusician

iMusician's terms exclude fully AI-generated content without a refund, defining it as content composed, arranged and produced by AI without a meaningful human contribution. AI used for inspiration, arrangement or sound design within a process led by a human remains permitted. Its quality assurance team reviews submissions case by case, with no published detection method. iMusician is also the distributor that has written most openly about how to challenge a platform's AI label. It advises artists to contact their distributor first, while warning that the final decision rests with the platform and cannot be overridden by the distributor. It lists the evidence platforms may request, including DAW project files, stems, session exports and dated drafts. Its article of 11/08/2026 does not, however, describe any specific case that has been resolved.

Sources:

LabelGrid

LabelGrid accepts AI-assisted and AI-generated music, provided that every release carries a mandatory disclosure. It covers three elements, the artwork, the sound recording and the composition (melody, structure and lyrics), each rated on the same four-level scale: none, minor, material, entirely AI-generated. The disclosure is sent to the streaming platforms. Tracks declared entirely AI-generated are not delivered to YouTube Content ID, YouTube Music, Meta, TikTok or SoundCloud monetisation; the rest of the distribution goes ahead and the release is not rejected. AI voices imitating real artists are not allowed.

The disclosure is checked during the review that precedes distribution, and LabelGrid does not publish the methods it uses. An honest disclosure does not get a release rejected, while a missing one gets it sent back to the artist, who can fix it and resubmit. A rejected release is locked, so the appeal goes through the support request form, sent from the email address on the LabelGrid account. It must include the release ID, the rejection reason and proof of rights (licences, evidence of master ownership or documents showing that the work is the artist's own). The team then reviews the release again, with no stated response time. LabelGrid publishes neither the share of fully AI-generated tracks it receives nor the share it rejects or removes on that ground.

Sources:

Ditto Music

Ditto Music accepts music made with AI tools as long as it complies with the platforms' guidelines, which means holding all the rights, including to the AI-generated material, not imitating another artist's voice or persona without permission and not flooding services with bulk uploads. Its help page, updated on 04/08/2026, warns that several services may reject fully AI-generated releases, naming Apple Music, Shazam, Beatport, iTunes, Audiomack, Meta Rights Manager, Pandora and YouTube Content ID, but says nothing about any detection or declaration of Ditto's own for distribution. Its publishing rules are clearer. Because collecting societies, according to Ditto, do not accept copyright claims for fully AI-generated songs, tracks that "our system" identifies as such cannot be submitted to its publishing service, and neither the system, its accuracy nor any way of challenging its verdict is described. Separately, on 22/05/2026 Ditto launched a licensing programme through which artists who wish to, and only with their explicit consent, can make their music available to approved AI partners for model training in return for payment.

Sources:

IDOL

IDOL's statement on AI, published on 22/11/2024 according to the page's metadata (no date is displayed), deals only with model training. The distributor acknowledges that responsibly produced AI tools have "significant potential for the music sector", provided that the rights of everyone involved in creating and distributing music are respected and licensed in advance. It then expressly reserves its rights, prohibiting the use of works and content it owns or controls for machine learning or AI training without prior written permission. The statement says nothing, however, about distributing AI-generated tracks, whether on accepting them, identifying them, labelling them or challenging a rejection.

Sources:

ONErpm

Since 28/02/2026, ONErpm has required artists to state at upload whether generative AI was used to create a track, placing it in one of four categories. In a human-created track with no AI, lyrics, melody and composition come directly from the artist. In an AI-assisted track, AI may suggest ideas, analyse sound or help with mixing, mastering or noise reduction. In a partially AI-generated track, AI generates parts of the music or lyrics, which the artist then edits, develops, selects or arranges. A fully AI-generated track is created mainly through prompts, with minimal artistic input. No other distributor in this table offers such a detailed scale. ONErpm says the declaration is "informational only" and that all tracks currently remain eligible for monetisation, with royalties unaffected, while encouraging honest declarations as preparation for future policy changes. The announcement does not say whether the declaration is passed on to platforms, and describes no detection, verification or appeal procedure.

Sources:

RouteNote

RouteNote accepts music created with AI under a few conditions, which its help page of 17/08/2026 sets out as guidance: check the terms of the sites you use, avoid imitating real artists and favour tools whose models are trained ethically. Its moderation team may ask for links to the tools used and warns that it may not approve content whose origin it does not know. Releases containing AI-generated or AI-assisted content may also undergo additional review and take longer than usual. No AI-generated content is delivered to RouteNote's content recognition services, to its Korean partner stores (Melon, Genie, Bugs, Flo and Vibe) or to Amazon. The page does not say whether this exclusion reflects a requirement from Amazon, which has published no AI music policy, or the distributor's own choice, and it documents no automated detection, accuracy figures or appeal procedure.

Sources:

Wiseband

In an article of 03/01/2024 on the reasons a release may be rejected, updated on 24/09/2026, Wiseband includes a section on content suspected of being AI-generated. Such content may enhance creation, the distributor writes, but it is also used to distribute automatically generated material designed to produce fake streams, and it may infringe protected works. Wiseband therefore says it prefers "to promote real music composed by real humans" and reserves "the right to reject this type of content at our discretion", without explaining how it identifies such content. The same article warns that "no additional answers will be provided" about rejections and that it will be "pointless and futile" to contact the company on the subject, an instruction that applies to all rejections, not only those involving AI. An artist whose release is turned down on these grounds therefore has no documented route of appeal.

Sources:

The table's blind spots

No platform has published an independent audit, so there is no way to tell whether the reported accuracy rates would withstand independent scrutiny. The only available error rates come from the platforms themselves. The treatment of hybrid creation is just as uncertain, since only Traxsource labels it and no one defines the threshold between "assisted" and "generated". Nor do we know how long an appeal takes or what evidence is needed for a decision, since no platform publishes that information. If you have been through one of these procedures, your case can be brought to public attention through the observatory of disputed decisions.

Updated 21/10/2026