Investigation · Credits
Where credits actually live
Six registers hold a song's credits. None of them is authoritative for the whole question, and almost every credit argument is two people quoting different ones.
Hip-hop argues about credits constantly and rarely specifies which record is being argued about. There are at least six, they were built for different purposes, and the ones that carry financial force contain no production credits at all, while the ones people actually read carry no financial force.
The six registers
| Register | Authoritative for | Cannot tell you |
|---|---|---|
| Streaming display | Nothing — it renders what the label supplied | Shares, splits, or who is missing |
| Label / distributor metadata | What reaches every platform | Whether the composition was registered anywhere |
| MLC works database | Registered composition shares (US digital mechanicals) | Who performed or produced |
| PRO registration | Who receives performance royalties | Mechanical shares, session personnel |
| Genius / Wikipedia | Nothing formally | Anything with legal or financial force |
| Liner notes | What the label chose to state at release | Anything that changed after, or was omitted |
The split that causes the arguments
Notice the shape of that table. The MLC and the PROs are the registers with money attached — and neither contains a single production or performance credit. They are composition registers. They know who owns shares of the song; they do not know who played on it, who engineered it, or who was in the room.
Meanwhile the registers that do list producers, engineers and features — the streaming display, Genius, the liner notes — have no financial force whatsoever. Being listed there pays nobody.
Why they drift apart
Spotify itself states that some label-provided credits are incomplete or may contain inaccuracies, and from 2025 it began rejecting releases at quality control that supply initials, nicknames or incomplete contributor information. That is a real improvement and it is also an admission of the prior state.
Underneath sits what the trade press calls metadata debt: the accumulated backlog of inconsistent supply-chain information. A producer credited under three name variants across five releases is not one contributor in the system. They are three, none of whom is fully paid.
The financial consequence is not marginal. Trade estimates put roughly a quarter of mechanical royalties due to publishers and songwriters as going unpaid because the rights cannot be identified — and the industry has been publicly promising complete streaming credits for more than a decade.
Why an assistant cannot answer this
Ask a general-purpose AI who wrote a song and you will get a confident answer. There is a documented case of the same question being asked ten times and returning ten different wrong answers.
The reason is structural rather than a bug: the crowd register is the one most heavily represented in the text these systems were trained on. Genius annotations, fan wikis and forum threads are abundant, well-formed and freely scraped. Statutory registration data is none of those things. So an assistant inherits crowd-level reliability while presenting it with uniform confidence, and it has no mechanism to tell you which register its answer came from.
That is precisely the gap a lookup can close, and why the credits question is worth building for rather than writing about.
What to actually check, in order
- For money: the MLC for composition shares, and your PRO for performance. If you are not in both, you are not paid from both. Being in one does not put you in the other.
- For the public record: the streaming display, and if it is wrong, the fix routes through your label or distributor rather than the platform.
- For history: the liner notes, understanding they were never a rights register.
- For culture: Genius and the fan record, understanding they carry no force and were never meant to.
What would prove us wrong
- A single register becoming authoritative across all four questions — ownership, performance, production and display. Nothing currently proposes to be that.
- A measured audit showing the registers agree in practice despite being designed differently. We did not measure agreement at scale, and if they largely concur, this framing overstates the problem.
- Crowd registers proving as reliable as registration data on a systematic comparison, which would remove most of the force of the assistant argument.
Limits, plainly. This describes how these registers are designed to work; it is not a measurement of their accuracy. We did not audit credit records against each other at scale. The ~25% figure is a trade-press industry estimate, not our measurement, and estimates in this area vary widely. Practice differs by territory, label and era, and older catalogue behaves differently from a current release.
The data
All six registers, compared on what they are authoritative for, who can edit them, what they structurally cannot tell you and what it costs when they are wrong.
Download data.json The credits beat
Collected 2026-08-12 · CC-BY 4.0, commercial reuse permitted, attribution and a link back required.
Referenced: The MLC ·
Spotify song credits ·
Spotify for Artists on metadata ·
WMOT, studio musicians and streaming credits
Related: The MLC ·
PRO ·
ISWC ·
ISRC ·
Split sheet ·
Who answers the money questions