DistroKid vs TuneCore vs CD Baby (2026): who actually accepts AI-generated music

Every comparison of these three distributors written before this spring is wrong about the only question an AI musician cares about. The old framing — DistroKid screens hardest, CD Baby is the soft touch — has inverted, because two of the three stopped asking about your track and started asking about the model that made it. We read all three help centres at source in October 2026 rather than trusting the secondary write-ups, and the distrokid vs tunecore comparison that falls out looks nothing like the one in most search results.

The 2026 answer in one table

Where the eight distributors we track stand, with the date each policy was last touched at source, re-read 28 October 2026. "Disclose" means the distributor gives you a field to declare AI and expects you to use it.

Two things jump out. Six of the eight accept AI music in some form, and the word "banned" appears in none of these policies — the belief that distributors have shut the door on AI is not what the documents say. And the two strictest are strict about something you may not be able to answer: what your generator was trained on. CD Baby is the exception to both at once, for reasons worth a section of their own.

A second trap sits in the disclosure column. UnitedMasters warns that services "may automatically flag or label your tracks"; DistroKid invites you to declare AI in structured credits; RouteNote instructs you not to reference the AI company in your metadata. Three distributors, three incompatible instructions, no way to satisfy all at once. There is no portable disclosure practice yet — only your distributor's current page. And LANDR's exclusions matter more than its cap: AI releases are dropped from Content ID, Meta, TikTok, Deezer and Pandora.

Can I upload AI-generated music to DistroKid?

Yes — the one question in this comparison with a clean answer, published under almost exactly that heading. DistroKid's help centre article Can I Upload Music Made With AI Tools to DistroKid? opens with one sentence:

Yes—DistroKid accepts music created with AI tools, but there are some rules.

The rules are four, and none is about AI as such: you must own 100% of the rights, "including the legal right to distribute music created with any AI tools"; no impersonation of "someone else's voice, likeness, or identity without permission"; no mass-generated spam "created solely to game streaming algorithms"; and no infringement.

Then the caveat that matters more than people expect: "streaming services may reject or remove releases that don't meet their guidelines." DistroKid accepting your upload is not Spotify accepting it. Note what is absent: no clause about which generator you used, no licensed-model requirement, no cap. Make a track in Suno, own the output, and DistroKid's published position is that you may upload it.

Alongside that sits AI Credits, the disclosure step DistroKid has carried since April 2026. You add them if AI generated part of the track — audio, lyrics, or melody and arrangement — and not if AI was only a tool: pitch correction, auto-tune, AI-assisted mixing, mastering or workflows. Many producers assume any AI in the chain triggers a declaration. It does not.

Where it surfaces has just widened: the AI Credits page, updated 20 October 2026, now reads "Streaming services (currently Spotify, Apple Music, and YouTube) show this info to listeners". YouTube was not on that list earlier in the year. Answer "all of the audio" at upload and it asks one more question — whether your artist name is a human or an AI persona.

Be blunt about the next part, because the confusion sells bad advice. AI Credits describe the recording. Spotify's AI Persona badge describes the artist identity. Different systems, different triggers, and neither is something a plugin or a re-render removes: what you declare on delivery is metadata, and metadata is not in the audio. Our Spotify AI music detection breakdown covers the downstream half.

TuneCore moved the question from your track to your model

TuneCore's GenAI Music Content Framework — last updated on 28 October 2026, so it is being actively maintained — states the rule in one line:

TuneCore only distributes music created using GenAI models trained on fully licensed datasets. Music created using models that rely on unlicensed datasets is not eligible for distribution through TuneCore.

Three details decide most real cases. It applies at any depth — "If GenAI is used at any point in the creation of a track, the tools involved must rely on fully licensed datasets" — so there is no AI-assisted carve-out of the kind DistroKid grants for mixing and mastering. The burden is on you: artists must review a tool's "documentation, terms of service, or licensing disclosures". And the approved list is one name long: Google Flow Music.

This is not abstract. On 30 April 2026 TuneCore's parent company Believe announced it would block tracks from unlicensed generators at upload, with Believe's CEO publicly pressing streaming services to do the same. It is enforced at the upload step rather than discovered later.

There is a second TuneCore gate that no comparison we found mentions, because it sits on the publishing side rather than distribution. A separate support article, GenAI and TuneCore Publishing, Explained, updated 27 October 2026, says the works TuneCore Publishing represents "must be substantially created by human songwriters", because compositions generated entirely by AI "cannot be registered with PROs/CMOs". How does TuneCore know? It says plainly: "We have implemented AI detection technologies across the catalogue we administer," and songs flagged as likely AI-generated "may not be processed".

Read together, the result is sharper than either document alone: a fully generated track made with an approved licensed model could clear TuneCore's distribution test and still be ineligible for publishing registration and the royalties that follow. Distribution asks what the model was trained on; publishing asks whether a human wrote the song.

CD Baby publishes two AI policies, and they contradict each other

Search for CD Baby's AI policy and you will be told, confidently and repeatedly, that it bans all AI-generated music. Last time we updated this page we called that a myth. It is not a myth — it is one of two answers CD Baby publishes, and the other says something different.

Start with the governing document, the Downtown Music content policy covering CD Baby, FUGA and Downtown Artist and Label Services:

Unlicensed AI Content: Content created wholly or primarily by AI models trained on copyright protected sound recordings without a license from the applicable copyright owners is prohibited. AI-generated content that mimics popular artists is also not permitted. This includes, but is not limited to, voice, image and likeness.

The qualifiers are the whole clause. "Wholly or primarily" leaves genuinely AI-assisted work outside it, and "without a license" makes this a training-data test — structurally the one TuneCore applies, not a ban on the technology. On its own it puts CD Baby and TuneCore in the same place by different routes.

Except that CD Baby's own help centre answers the same question in an article titled Do I need to own all the rights to the music I'm selling?, and answers it the other way:

Artificial Intelligence — You will not be able to distribute A.I.-generated content. Even if the A.I. generator you are using permits commercial use of what you made, there's no way to ensure unique sounds or rights compliance. For this reason, CD Baby can't distribute this type of content.

No "wholly or primarily". No licensing qualifier. We checked both on 28 October 2026 and both were live, three days apart in their last update. One describes a training-data test the other does not acknowledge exists.

We cannot tell you which one an inspector applies, and neither can CD Baby's help centre. Assume the stricter sentence is operative, because it is the one that costs you a $9.99 submission and a fortnight if it turns out to be right. On a licensed-model track, the honest answer is that CD Baby's published rules do not agree with each other.

The pricing has moved too

We re-checked all three at source on 28 October 2026. If you are reading older guides, assume their numbers are wrong.

Three corrections, all re-read at source on 28 October 2026. DistroKid's Ultimate at $89.99 covers up to 100 artist names, not the five of the old "Label" plan. TuneCore's per-release album renews at $56.49 after the first year, so the $44.99 is a first-year number. And CD Baby's 9% is not buyable — the $39.99 Boost add-on adds sync, SoundExchange and MLC collection, but takes its own cut of each.

Release twice a year or more and DistroKid is cheaper than CD Baby before you reach the commission. Release once, ever, and CD Baby's $9.99 wins on day one and slowly gives it back.

"Lenient" is three different questions

The most-asked version of this comparison is which distributor is easiest, and the people asking are more precise than the guides answering them. One put it exactly: "I don't mean just speed, but how lenient the review process is." That is two questions — and a third matters more than either: what happens to a release already approved.

Read at source from each distributor's help centre on 28 October 2026: DistroKid delivery times, TuneCore release statuses, CD Baby processing times.

Three things fall out. The speed gap is not marginal — a fortnight of CD Baby inspection against two business days at DistroKid is the difference between hitting a release date and missing it, and CD Baby charges $29.99 to close it. TuneCore's rejection route is the gentlest on paper: a Denied status and an email thread, where DistroKid asks you to delete and start again. And the one nobody budgets for: approval is not permanent at any of them.

Which stores you can turn off, and how hard it is to leave

The second thing people switching distributor ask about is control — in one poster's words, "control over what platforms/services/companies songs are sent to". All three let you pick stores at upload; they differ sharply on reversing that later, and no competitor on this search result builds the column.

The asymmetry is the finding. CD Baby gives the most self-serve control and the least room to change your mind — pull a single platform yourself, but the help centre says plainly that it is permanent. DistroKid is the mirror image: leaving one store needs a human, yet nothing in its documentation treats removal as irreversible. TuneCore is strictest on the way out — "We can't reverse takedowns, as stores consider this action permanent." Moving a catalogue rather than a single, that ordering matters more than the annual fee.

Where a Suno track actually stands in late 2026

The question underneath the question, and the answer is genuinely uncertain — so here is the state of play rather than a verdict. Warner Music settled with Suno in November 2025, under a deal in which Suno committed to deprecating models trained on unlicensed music. Universal and Sony did not settle.

That fight now reaches the distributor directly, which is why people write that DistroKid "used to let almost everything through" in the past tense. On 14 October 2026 UMG and Sony asked a Manhattan federal court to compel DistroKid to comply with a subpoena, seeking monthly counts of the AI-generated tracks it delivers. The detail that matters here is incidental to the case: according to the filing as reported, DistroKid uses detection technology to identify AI-generated submissions. DistroKid disputes the data's relevance and says it remains committed to cooperating. artifactr covers the case in UMG v DistroKid; the narrow point here is that a distributor whose published policy asks only about your rights still runs AI detection on what you upload.

Under a training-data test, meanwhile, a Suno track's eligibility at TuneCore or CD Baby depends on which model generated it — and most users cannot answer that about a track they made last year.

None of that is the same problem as screening, and conflating the two is why people buy the wrong fix. A screening rejection is about the file — audible generation artifacts and generator signatures that automated checks flag before a human listens. That is a file problem, and Undetectr exists for that step. It will not change a platform's AI label, your disclosure obligations, or whether your model was licensed — nothing will. It handles the screening gate and stops there. Our pillar on how distributors detect AI music walks through both gates in order, and eraseai.co covers the DistroKid declaration fields.

So which one should you actually use?

Simpler than the policy reading suggests, because in practice your generator picks your distributor.

Suno, Udio, or any generator whose training data is contested → DistroKid. The only one of the three whose test you can pass by yourself, cheapest per release above about two releases a year, and reviewed in two business days.

Google Flow Music, or any generator with published licensing → TuneCore, whose framework welcomes exactly this case. Accept that a takedown is one-way and publishing asks a second question.

One release, no recurring fee, time to spare → CD Baby at $9.99 — but budget for the 9% forever, the 7-14 day inspection, and the unresolved contradiction in its own AI policy.

The part nobody wants to hear

Getting distributed stopped being the hard part a while ago. The AI music subreddits keep saying the same thing: distribution is solved and nobody can find an audience — a scrape of the main AI-music hashtags in September 2026 found a top post at 310 likes and most under ten. Picking correctly between these three gets you onto the shelf. It does not get you heard.

So the more useful question after release is where money comes from without algorithmic discovery. Paid sync placements in TV, film, games and ads are where the money conversation in this niche actually happens, and played.fm is built around that route plus selling direct and keeping the proceeds. Worth knowing the comparison point: Bandcamp bans music made wholly or substantially by AI, so it is not an option here. Neither route is a growth strategy on its own, but neither depends on being recommended.

Pick the distributor whose test you can pass, disclose the way it asks, and treat screening as the separate, solvable file problem it is.