The six-point chart test for AI music
Eleven labels proposed six conditions for AI chart eligibility and no chart adopted them. Turned into a pre-release checklist a real release can pass now.
On 29 July 2026, eleven labels put out a joint proposal for how charts should treat AI-involved music: Universal, Sony, Warner, Believe, BMG, Concord, Dirty Hit, Glassnote, HYBE, Mom+Pop and Partisan, reported by Billboard. Six conditions, all six required, or the track does not count.
Neither Billboard nor the Official Charts Company had adopted them as of 30 July. The proposal is a proposal. The trigger, though, was real: the synthetic act IngaRose topped both the US and global iTunes charts in April 2026, and a chart that cannot distinguish that from a signed artist's release has an obvious problem to solve.
Here is the useful way to read a six-point test that nobody is enforcing. Five of the six are things you would want documented anyway, for reasons that have nothing to do with charts. The sixth is genuinely hard. Treating the whole list as a pre-release checklist costs you an afternoon and produces a file you will be glad to have when a distributor, a DSP, a sync agent or a platform asks a version of the same question, which they will, on their own timeline, without warning.
The six conditions
| Condition | What it asks | Who can verify it |
|---|---|---|
| Substantially human-made | A human contribution material enough to count as authorship | Nobody, yet — no threshold is defined |
| Built on properly licensed AI | The tools used were trained on lawfully obtained material | The tool vendor, via its own disclosures |
| Free of stream and chart manipulation | No bought streams, no bot farms, no coordinated inauthentic play | The DSP and the chart compiler |
| Copyright compliant | No infringing use of existing recordings or compositions | You, and a claim system after the fact |
| Personality-rights compliant | No unlicensed use of a real person's voice, name or likeness | You, and the person concerned |
| Consistent with the AI service's terms | The output was produced within the tool's own rules | The vendor, from your account record |
| Disclosed as AI-involved | The AI contribution is declared, not concealed | You, at submission |
That is seven rows for six conditions because copyright and personality rights are named together in the proposal and pull apart cleanly in practice. Keep them separate on your own checklist; they fail in different ways and are fixed by different people.
"Not adopted" is a bad reason to ignore it
Three reasons this list is worth adopting unilaterally.
It is the convergent shape. Look at what platforms have already done. Streaming services moved to labelling AI involvement rather than removing it, which is the disclosure condition arriving under a different name and covered in tagged, not removed. Video platforms have their own version aimed at mass-produced uploads. Licensing questions are being asked of AI vendors by their own partners. The six conditions are not eleven labels inventing a standard from nothing; they are eleven labels writing down the standard everyone is separately converging on.
Volume makes gatekeeping inevitable. When the supply of new tracks is effectively unbounded, every downstream system has to develop an admission test, because the alternative is drowning. The economics of that are the subject of ninety thousand tracks a day. A test that does not exist yet will exist, and the version that ships tends to look like the version that was proposed.
The evidence is cheap now and expensive later. Every item below is trivial to capture at the moment you make the track and near-impossible to reconstruct eighteen months on. That asymmetry is the whole argument.
The checklist a real release can pass today
Run this before the release, not after. Each item names the artifact you keep, because "we were careful" is not evidence.
Log the tools, per track. Model name, version, date, account, and what each was used for: top line, arrangement, mastering, stem cleanup. One row per tool per track. This single file answers three of the six conditions on its own.
Keep the vendor's terms as they stood on the generation date. Terms change. Save the PDF or the archived page with the date. The "consistent with the AI service's terms" condition is asking whether you complied with the rules that existed when you generated, and only a dated copy can show that.
Confirm the plan grant covers commercial release and distribution. Commercial use and distribution to DSPs are frequently separate grants. Read the clause against what you are actually doing. The general landscape is in the AI music licensing guide.
Write the lyrics yourself, and keep the drafts. Dated drafts with revisions are the cleanest evidence of human authorship anyone has, and keeping them costs nothing but a folder. If you use a model to break a block, keep that session too and keep your rewrite next to it. The agent's lyric step is deliberately text-only for this reason: you get words to review and edit before any audio is generated, which leaves a paper trail where a single one-shot generation leaves none.
Record one real element. A vocal take, a guitar part, a piano line, a percussion overdub. One is enough to change the character of the "substantially human-made" argument from a description to a fact, and it usually improves the record too.
Document the arrangement decisions. A short note per track: what you cut, what you re-ordered, what you re-recorded, which take you chose and why. Editing choices are authorship, and they are invisible in the master unless you write them down. Stem separation is where a lot of this work happens, and stems make the before-and-after legible.
Clear every voice that resembles a person. If the vocal was cloned, modelled on, or prompted toward a named artist, you need a signed permission or a different vocal. Personality rights are the condition with the least ambiguity and the sharpest consequences.
Run a similarity check before release. Play the top line against the obvious reference. If the melody is recognisably something else, fix it now. Content-ID systems and human A&R ears both find this eventually.
Do not buy a single stream. Obvious, and yet: chart manipulation is the one condition on the list that is already fully enforced by everyone, with existing detection and existing penalties. Playlist services promising guaranteed placement fail this on your behalf.
Write the disclosure line before submission. Distributors ask; write it once, use it everywhere, keep it factual. "Vocals performed by the artist. Instrumental arrangement generated with [tool, version] and edited by the artist" is worth more than "made with AI" because it is specific enough to be checkable.
The one you will argue about
"Substantially human-made" has no defined threshold, and nobody has offered one. That is not a drafting oversight so much as an unsolved problem: percentage-of-the-waveform tests are meaningless, and time-spent tests reward inefficiency.
What works in practice is making the human contribution legible rather than trying to make it large. Three contributions are legible without argument:
- The words, when you wrote and revised them, with drafts.
- A performance, when a human played or sang something that is audibly in the final master.
- The selection and arrangement, when you can describe which options you rejected and why.
Any two of those, documented, produce a much stronger position than a one-shot generation that happened to take you an hour of prompting. The distinction the labels are groping toward is the same one video platforms are drawing: it is not about how much machine involvement there was, it is about whether a person made decisions that show up in the result.
For artists thinking about the release around the record rather than the record itself, the campaign side is covered in AI video for musicians and album launches, and the smaller-scale version in indie musician video marketing.
FAQ
Do I have to do any of this if the charts have not adopted the rules?
For chart purposes, no. For everything else, the same questions arrive from distributors, sync agents, ad platforms and DSPs on their own schedules, and the answers all come from the same file. The checklist is worth running because it is portable, not because a chart demands it.
Does using AI at all make a release ineligible under the proposal?
No. Every one of the six conditions is about how the AI was used and whether it was declared, not whether it was used. A track with a generated arrangement, a human-written lyric, a real vocal take and a clear disclosure is aiming squarely at what the proposal describes as acceptable.
What counts as disclosure, exactly?
Nobody has specified a format, which means specific beats vague. Name what was generated and what was performed, at the point of submission and in the release metadata. A line that describes the actual split is harder to challenge than a blanket "AI-assisted" tag, and it is the same line you will want on the video assets around the release.
Is any of this enforceable against an anonymous synthetic act?
Not really, and that is the honest weakness of the whole approach. A checklist rewards people with something to lose. What it can do is make the difference between a documented release and an undocumented one visible to the systems downstream, which is a different lever than enforcement and probably the one that ends up mattering.