⏳ Stanford: honest AI labels cost listeners

Issue #2 · Friday, May 15, 2026 · ~7 min read

AI disclosure hits the upload page

DistroKid, the largest independent distributor in the world, added an AI disclosure toggle to its upload flow on May 14. The same day, Spotify expanded its AI Credits beta into more song metadata. Both moves land in the same week Stanford published peer-reviewed evidence the label depresses engagement even when the song is human-made.

• Apple Music's Transparency Tags went live in March 2026. 33% of new uploads now arrive self-reported as fully AI-generated.

• Deezer's automated detection flags around 75,000 AI tracks per day, roughly 44% of new uploads.

• DDEX published its AI Disclosure Standard last quarter, ahead of the EU AI Act's August 2 enforcement deadline.

This week, disclosure moved from the listener-facing layer to the upload layer.

DistroKid now asks every uploader to mark releases as AI-generated or AI-assisted, with the toggle baked into the standard release page. Spotify's AI Credits expansion does the same job downstream by surfacing the disclosed status inside track metadata. Both rely entirely on self-reporting.

The architecture is now visible. Apple Music holds the listener-facing tag. Spotify holds the metadata credit. DistroKid holds the upload-level declaration. Deezer holds the automated detection layer the others haven't built. The next 18 months decide which of those four positions becomes load-bearing once the EU AI Act takes effect and self-reporting stops being optional.

In today's briefing:

• DistroKid and Spotify ship AI disclosure on the same day.

• BPI counts 274 AI licensing deals, 26 of them in music.

• Bloomberg calls Billboard "lax" on AI artists charting unchecked.

• Music Intelligence: Chartlex finds a 66x gap between AI uploads and AI listening.

HIGH SIGNAL NEWS:

[Policy] Bloomberg calls Billboard "lax" on AI artists charting unchecked.

Bloomberg opinion columnist Peter A. Berry pushed Billboard to license AI detection software, naming Xania Monet (44.4M U.S. streams, $3M Hallwood Media deal, No. 3 Hot Gospel Songs) and citing 10+ AI acts on Billboard rankings between August and November 2025. Berry pointed at Deezer's third-party detection tool as the obvious fix.

What this means for you → For A&R, this is the first major-outlet pressure on chart legitimacy. SIQA already runs a creator-attested classification at 81% meaningful human contribution. For creators using AI, expect "human contribution percentage" to start showing up in eligibility rules.

[Artists] Jack Antonoff calls AI music creators "godless whores" in an Instagram journal entry.

Grammy-winning producer Jack Antonoff (Taylor Swift's Midnights, Lana Del Rey's NFR) posted a handwritten note framing AI music creators as soulless replacements. The post hit around 10,000 likes ahead of his Bleachers album release on May 22. Antonoff lumps AI-assisted producers in with prompt-and-pray creators, collapsing the workflow distinction the rest of the week's data treats separately.

[Licensing] BPI counts 274 commercial AI licensing deals, 26 of them in music.

The BPI's new report tallies 274 named rightsholder / AI licensing agreements globally and 26 music-specific deals (Suno, Udio, Stability, Meta, TikTok, Google). 16% of surveyed UK indies are already exploring partnerships and 77% are open. 97% of companies say existing copyright frameworks are sufficient. The report exists to slow further UK legislation following March's opt-out collapse.
What this means for you → For labels, the BPI is making your market-readiness argument for you. For independent artists, "77% open to licensing" doesn't mean 77% capable. Indie metadata and negotiating leverage are still gaps the BPI report avoids costing out.

[Rights] Sonilo licenses Shutterstock's full music catalog for video-to-music AI training.

Sonilo signed a named licensing agreement with Shutterstock covering its Pond5 and PremiumBeat catalogs. Contributors get around 20% royalty on Shutterstock's take. Shutterstock's AI licensing revenue: $104M (2023) heading toward a $250M target by 2027. CISAC projects 60% of B2B library music will be AI-generated by 2028.
What this means for you → For composers earning from stock library subs, treat that income as a declining asset over a 3 to 5 year horizon and audit your contributor opt-out settings this weekend.

[Tools] Mozart AI ships Auto-Stretch to lock every clip to project BPM. 

Mozart AI added Auto-Stretch on May 12, locking every clip (generations, uploads, recordings) to project tempo using zplane's élastiquePro engine, the same DSP licensed inside Ableton Live, Serato, and Steinberg Cubase. The unglamorous workflow fix that makes AI stems session-ready instead of demo-grade.

DistroKid's checkbox meets Stanford's data

The lede. DistroKid added an AI disclosure toggle to its upload flow on May 14, asking every uploader to mark releases as AI-generated or AI-assisted. Spotify ran the same play downstream the same day, expanding its AI Credits beta into more song metadata. Both rely on the artist tagging themselves.

Why it matters. Disclosure has now reached the upload step at the largest indie distributor in the world. Pair that with Apple's Transparency Tags (live since March) and Deezer's automated detection (75,000 AI tracks flagged per day), and the streaming stack now has four working layers of AI labeling before the EU AI Act enforcement deadline on August 2. The question has shifted to whether self-reporting is the right base for the next 18 months.

The Stanford signal:

A peer-reviewed study from Sarah H. Wu (Stanford) and Kevin J. Holmes (Reed College), published May 13 with 399 U.S. respondents, found listeners engage less deeply with tracks labeled "AI-generated" regardless of whether the track is AI. The inverse holds: AI tracks falsely framed as human-made drew more engagement than honestly-labeled AI tracks.

• 399-respondent peer-reviewed sample.

• Effect is preconception-driven, not signal-driven.

• The label depresses engagement on cold listens with no artist story attached.

The Chartlex audit
A separate audit of 36,500 indie catalogs over 12 months found AI tracks underperformed human releases by 25 to 40% on saves and showed +30 to +50% skip rates. Deezer's earlier data: 85% of streams on labeled AI tracks are fraudulent.

The unanswered question:

Both data sets measure only fully AI-generated tracks. AI-assisted work (mastering, vocal tuning, stem separation, arrangement help) sits inside the "human" baseline. Nobody is measuring the hybrid layer, which is where most working producers ship. Until that gets measured, every disclosure regime will lump the prompt-and-pray crowd in with producers using AI as a workflow tool.

The takeaway for music industry pros:

For A&R, the engagement-depressing effect of the label is now empirical, not theoretical; expect catalog teams to start segmenting by AI-assisted vs fully AI-generated when sizing release plans. For artists using AI as part of their workflow, the cost of staying silent will rise as automated detection comes online behind the self-reporting layer.

MUSIC INTELLIGENCE

RESEARCH & DATA
Chartlex — 66x gap between AI uploads and AI listening time.

Apple Music: 33% of new uploads are fully AI-generated, less than 0.5% of listening time. Chartlex audited 36,500 indie catalogs and found AI tracks underperform humans by 25 to 40% on saves with skip rates up 30 to 50%. The single insight worth pulling: the report excludes AI-assisted tracks entirely, so the visible "AI" share is the floor, not the ceiling.

CHART WATCH Tencent Music Super VIP hits 20M+ subscribers in Q1 2026.

Tencent Music posted $1.15B in Q1 revenue (+7.3% YoY) with its Super VIP tier crossing 20M subscribers, up from 15M two quarters ago. The same earnings call disclosed 27,000 AI tracks removed for "song theft, song laundering, and trend hijacking" via audio fingerprinting and voiceprint detection. China is the only major market running the enable-and-enforce dual stack at scale.

TECH WATCH: Suno fights to keep its WMG settlement sealed as UMG and Sony push for discovery.

The case has shifted from damages to licensing-rate benchmarking. Whatever per-track or per-generation rate WMG accepted will anchor every AI music licensing tool built in the next decade. The GEMA Munich ruling on June 12 is the next pressure point. What licensed AI tools look like depends on what number gets unsealed.

DISCUSSION OF THE WEEK: Markos Koumoulas: AI music platforms fail non-Western instruments.

Ethnomusicologist Markos Koumoulas, PhD, tested Suno, Udio, ElevenLabs, and Google Lyria 3 on three non-Western instruments (didjeridu, zhongruan, kemenche). Zero platforms reproduced the actual instrument or its stylistic role. Outputs collapsed into a flattened "world music / new age / ethnic ambient" sound.

IDEA WORTH SITTING WITH: Artistic Futures — an EU project using AI to recover missing songwriter royalties.

A two-year EU consortium led by Unison Rights (Spain) with seven partners across six countries is using AI metadata-matching to recover unclaimed royalties. CISAC estimates 25% of global royalties go uncollected, around $2.5B+ a year. The same matching tech could also identify unlicensed AI training data. Dual use. Nobody is naming it yet.

Jack Antonoff called AI music creators "godless whores" on Instagram this week. The framing lumps the prompt-and-pray Spotify farmer in with the producer who used iZotope Ozone for mastering. I'd push back on the collapse.

Stanford (Wu, Holmes, 399 respondents) showed listeners disengage from anything tagged "AI-generated" even when the song is human-made. The label hurts because the song arrives with no story attached. Olivia B. Moore, China Styles, and Slime Dot use AI heavily, and listeners didn't bail when they found out: there was a face, a project, and a process to point at.

Apple Transparency Tags and Spotify AI Credits will surface this data whether artists want them to or not. I wrote a disclosure template you can paste into release notes: tool name, exact role, the human calls you made. A production credit reads differently than a checkbox.

If you're using AI heavily (full generation, not a voice clone of yourself), being upfront earns the listener who emails because they connected with the process. The label lands in August. The story around it is what you still control.

Audit your metadata before the matching systems do.

Open every distributor and collection society dashboard you use. Check ISRC codes, IPI numbers, songwriter splits, and publisher info across your catalog. Artistic Futures, the EU consortium that launched this week, is building AI metadata matching to recover unclaimed royalties (CISAC: around $2.5B a year goes uncollected from broken metadata). Clean metadata puts you first in line when those systems route money back. Skip if you audited in the last 90 days.