Industries / Media, arts and recreation / Media and entertainment / Music recording
Labels and distribution
Labels and distributors turn finished recordings into products: they choose the format, make the copies or data files, deliver them to shops and streaming services with the right details attached, and deal with piracy and the flood of uploads.
7
Milestones achieved, from 1898 to 2025.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1898
Discs become a product
Emile Berliner patented his flat-disc gramophone in 1887, and companies were formed to make and sell its records. William Barry Owen and Trevor Williams set up the Gramophone Company in London after a meeting in December 1897, and the Encyclopedia of Greater Philadelphia says "by 1898 Gramophone and disc sales were soaring". Eldridge Johnson incorporated the Victor Talking Machine Company in 1901.
Credit
Emile Berliner; William Barry Owen and Trevor Williams (Gramophone Company); Eldridge R. Johnson (Victor)
1948
The LP and the 45
In June 1948 Columbia Records released the first long-playing microgroove record, by the Library of Congress's account. It turned at 33⅓ revolutions per minute (rpm) and held about 23 minutes a side. On 10 January 1949 RCA Victor unveiled a rival 7-inch disc that played at 45 rpm. From then on, albums and singles were sold on different discs.
Lessoncredit for a format is often shared, and the Library of Congress names a team rather than one inventor.
Credit
Peter Goldmark's Columbia and CBS engineering team, with Bill Bachman; RCA Victor
1982
The CD goes on sale
Philips showed the compact disc (CD) to the press in Eindhoven in March 1979, and Philips and Sony agreed on a 12 cm disc. In 1982 Sony released the CDP-101 player in Japan, priced at 168,000 yen, by the Computer History Museum's account.
Credit
Philips and Sony
2001
Napster loses in court
Napster ran a peer-to-peer network, in which users copied MP3 music files straight from each other's computers. Record companies sued, and the US Court of Appeals for the Ninth Circuit held that this copying was not fair use. It accepted evidence that Napster cut CD sales and "hindered plaintiffs' ability to enter the digital sales market".
Lessonwinning in court stopped the free copying but did not give buyers a legal place to download music; that came from a store.
Credit
A&M Records and other record companies; the Ninth Circuit
2003
Songs sold one by one
On 28 April 2003 Apple opened the iTunes Music Store, selling songs at 99 US cents each with no subscription, from more than 200,000 songs from BMG, EMI, Sony Music, Universal and Warner. Buyers could burn songs to an unlimited number of CDs for personal use.
Credit
Apple
2006
Artists release without a label
TuneCore launched in January 2006, charging artists a flat fee, US$9.99 a song, to place music in online stores such as iTunes; others took a share of sales. By 2010 the Los Angeles Times reported it was sending out 15,000 to 30,000 new songs a week. In May 2006 record companies, rights societies and digital music services launched DDEX, a standards body that sets shared data formats, including the message that tells a store what a release contains.
Credit
Jeff Price (TuneCore); DDEX's founders
2025
Streams pay, uploads flood
By IFPI's count, recorded music revenue grew 6.4% to US$31.7 billion in 2025; streaming brought in over US$22 billion, 69.6% of the total, and physical formats grew 8.0%, with vinyl up 13.7%. By Luminate's count, an average of 106,000 new tracks reached streaming services each day in 2025, up from 99,000 in 2024, each counted by its ISRC, the code that identifies one recording. Of 253 million tracks on the services, 88% got fewer than 1,000 streams that year.
Lessonthe catalogue grows far faster than listening, so most releases now find almost no listeners.
Credit
IFPI and Luminate, who did the counting
Watch for
AI uploads pass half
Deezer says more than 50% of new uploads to its service were fully AI-generated by July 2026, about 90,000 tracks a day, up from 39%, or 60,000 a day, in January 2026. Deezer began tagging AI albums in June 2025 and keeps them out of its recommendations and editorial playlists. It says its detector finds "100% AI-generated music from generative models such as Suno and Udio", and it now licenses the detector to others. One single-author preprint from September 2026 (ArtifactBench) scored the public Deezer detector at 0.761 AUROC, a score where 1.0 means AI and human tracks are sorted perfectly.
Achieved whenan outside party tests the detector on music from current generators and publishes error rates that back the claim.
Claimed by
Deezer, January and July 2026
Artists approve their own releases
Music is sometimes delivered to the wrong artist's page: in July 2025 an AI-made track appeared on the Spotify page of Blaze Foley, who died in 1989, and in September 2025 Spotify said it had removed over 75 million "spammy tracks" in 12 months. Spotify launched Artist Profile Protection as an optional test on 15 March 2026 and opened it to all artists on 17 November 2026. It tells artists when music is delivered under their name and lets them remove themselves from releases that are not theirs; a stricter "Block releases" setting, still being tested with a limited group, holds releases until the artist approves. Spotify calls it "a first-of-its-kind solution".
Achieved whenan outside count shows fewer wrongly credited releases on Spotify after the rollout.
artists.spotify.comsupport.spotify.com404media.conewsroom.spotify.com
Claimed by
Spotify, March and November 2026
AI credits in release metadata
In September 2025 Spotify said it was helping build an industry standard, through DDEX, for declaring AI use in music credits, and named 19 distributors and label-service companies working with it, including DistroKid, CD Baby, Believe and FUGA. In April 2026 it began testing "AI Credits" in the Song Credits panel of its mobile app. Spotify itself says "the absence of a credit doesn't mean AI wasn't used", and that not all distributors let artists disclose yet.
Achieved whenDDEX publishes the standard and an outside count finds it used in deliveries from several distributors to several services.
Claimed by
Spotify, September 2025 and April 2026
Open
01
Count missing credits at release
A checker that takes a sample of newly released recordings and records, for each streaming service, whether the "core data set" is present and correct. That is the list of basic details the UK music industry agreed to supply at release: recording title, artist, ISRC, songwriter names, and the IPI and ISWC codes that identify songwriters and songs.
The briefCredit · unclaimed02
Measure the AI-disclosure gap
A public tally that compares the tracks an AI detector flags with the tracks that carry an AI disclosure in their credits.
The briefCredit · unclaimed03
Tally misattributed releases across services
A weekly check of the artist pages of a panel of consenting artists on several streaming services, logging any release under their name that they or their label did not make.
The briefCredit · unclaimed
Open 01 · the brief
Count missing credits at release
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A checker that takes a sample of newly released recordings and records, for each streaming service, whether the "core data set" is present and correct. That is the list of basic details the UK music industry agreed to supply at release: recording title, artist, ISRC, songwriter names, and the IPI and ISWC codes that identify songwriters and songs.
Why it is still open
In April 2026 the UK Intellectual Property Office (IPO) said metadata "is not as accurate or complete as it should be", and that the measures for tracking progress were still being tested; its page gives no completeness figures. The nearest count is a one-off: in April 2024 Brazil's collecting society UBC checked 100 songs on five services and found that 31% on average lacked credits or had errors.
Built with what exists
Python; MusicBrainz's public database downloads as a reference; the services' public pages or developer interfaces, used within their terms; Gemini Decision 1 to sort credit records in bulk (see the Note "A thousand questions, one pass"); Claude to review odd cases (see the Note "Seven more flagged").
First step
Pick 100 recordings released in one week, record by hand the credits each of four services shows, and compare them with MusicBrainz. That hand tally becomes the checker's test set.
Done looks like
A published dataset of at least 1,000 recordings released in one month, checked on at least four services, with a completeness rate for each core field and service, a 100-item hand-checked subsample showing the checker's own error rate, and the same run repeated three months later.
Open 02 · the brief
Measure the AI-disclosure gap
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public tally that compares the tracks an AI detector flags with the tracks that carry an AI disclosure in their credits.
Why it is still open
Spotify's AI credits depend on artists choosing to disclose, and Spotify says a missing credit "doesn't mean AI wasn't used". No study measuring how often AI tracks are actually disclosed was found.
Built with what exists
Two 2026 releases make this possible now: Deezer's free online detector, launched on 11 June 2026, which scans playlists from about 20 streaming services, and Spotify's AI Credits panel, in testing since April 2026. Also a spreadsheet, and a browser agent such as Browser Use for repeat checks where the service's terms allow it (see the Note "Deleting 80% to start over").
First step
Build one public playlist of 100 releases from this week, run Deezer's detector on it, and check the Song Credits of every flagged track by hand.
Done looks like
500 new releases from public new-release playlists, each with the detector's verdict and its disclosure status, and a published share of flagged tracks that carry an AI credit. Report the detector's known error rate beside the result, re-check 50 tracks by hand to measure recording errors, and download no audio.
Open 03 · the brief
Tally misattributed releases across services
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A weekly check of the artist pages of a panel of consenting artists on several streaming services, logging any release under their name that they or their label did not make.
Why it is still open
Spotify's tool covers only Spotify, and Spotify itself says music "has been landing on the wrong artist pages across streaming services". No outside count of how often this happens was found; the figures that exist are company counts.
Built with what exists
MusicBrainz as the reference list of each artist's releases; the services' public artist pages or developer interfaces, within their terms; a browser agent for the weekly snapshot (see the Note "Deleting 80% to start over"); a small database.
First step
Recruit 10 artists, for example through a college music programme, and take the first snapshot by hand.
Done looks like
12 weekly snapshots for 50 artists on four services, each flagged release confirmed or rejected by the artist or label, and a published tally by service, so Spotify's rate after November 2026 can be compared with the others.