9 Signals That Matter | Week of September 25, 2026

What you need to know in the music business this week

1. LEGAL WATCH

Artists Are Asking Where Their AI Money Is. Labels Still Haven't Answered.

A Los Angeles Times investigation this week put a specific shape on something the music industry has been dancing around since the first major AI licensing deals were announced.

Warner, BMG, Believe, Universal, and others have struck AI licensing agreements that could eventually generate significant revenue. The financial mechanics for the artists whose work makes those deals valuable remain remarkably opaque. The companies haven't publicly disclosed how much participating artists receive, how many have opted in, or exactly what control individual artists have over training and downstream uses of their recordings.

The American Federation of Musicians is suing Universal and Warner, alleging that recordings containing union performers were licensed for AI use without the compensation and notification required under their collective bargaining agreement. That case is working through the courts alongside the broader copyright litigation.

But the contractual wrinkle that matters most for anyone negotiating agreements right now goes beyond the AFM case.

An artist may theoretically have an opt-in right for AI uses of their recordings. Whether that right is meaningful depends on who actually has authority to exercise it. And in most recording agreements, the label controls the master. The artist's nominal right to opt in may be subject to the label's override in ways that weren't contemplated when the agreement was written, because nobody was contemplating AI training as a use case when most of those agreements were signed.

AI rights are becoming a new due-diligence category that belongs in every artist and label agreement from this point forward. The questions worth demanding explicit answers to: who can authorize AI training on these recordings, who can authorize voice and likeness simulation, is consent use-specific or blanket, can it be revoked, and where exactly does the money go when those rights generate revenue.

The majors have established that AI rights have value. Artists need to make sure they haven't accidentally signed those rights away in agreements that predate the question.

2. CULTURAL SIGNAL

An AI-Assisted Song Charted in Korea. The Biggest Demographic Was People in Their 40s.

"Forest of the Heart," written by a human but composed and vocalized by Suno, became the first generative-AI-assisted recording to enter Melon's Top 100, one of South Korea's most influential streaming charts. It peaked at number 89. The largest listening demographic was people in their 40s, at 38 percent of the audience. It spread largely by word of mouth, not algorithmic promotion.

Listeners apparently didn't care how it was made. They just liked the song.

That's the uncomfortable but important data point the music industry needs to sit with and stop avoiding.

The industry spends enormous energy debating whether consumers will accept AI music. It holds symposiums about authenticity, commissions studies about listener preferences, and issues policy statements about human creativity. Meanwhile, an AI-assisted record cracked a major chart because middle-aged Korean listeners found something in it they wanted to share with other people.

Consumers don't shop for production methodology. They shop for something they like. If an AI-assisted record connects emotionally, a meaningful percentage of listeners will not reject it solely because AI touched the process. The evidence is now in the chart data, not just in theoretical arguments.

That means provenance, disclosure, and rights administration have to do the heavy lifting that consumer rejection was assumed to provide. The industry cannot count on audiences self-selecting away from AI music based on how it was made. It has to build the infrastructure that distinguishes, labels, and fairly compensates the full spectrum of human-to-AI creative involvement.

The market isn't going to solve this. The chart already told us that.

3. MARKETING

Podcast Advertising Is Posting 4.2x ROI. Here's Why That Number Makes Sense.

A new Audacy Podcasts report cites econometric research from Acast and Annalect finding that podcast advertising generated an average 4.2x return on investment, compared with 3.6x for social media, 3.2x for display advertising, and 2.2x for search.

Apply appropriate skepticism to any vendor-sponsored ROI study. The methodology matters, the sample matters, and numbers like these get circulated because they're favorable to the people circulating them.

But the underlying behavioral argument that produces a number like 4.2x holds up independently of any single study. Podcast advertising combines three things that most social advertising doesn't: sustained attention from a listener who chose to be there, host trust built over repeated interactions, and contextual relevance because the listener self-selected into a topic that intersects with their actual interests.

Compare that to a social ad that interrupts someone who was trying to look at something else.

For artist marketing, the experiment worth running isn't buying generic prerecorded spots against broad audience targeting. Find podcasts where the artist naturally belongs: regional music shows, genre-specific programming, lifestyle and affinity content where the artist's audience already spends time. Then test host-read endorsements, interviews, and integrated segments rather than standard ad units.

A credible host telling 15,000 highly aligned listeners why they personally responded to a record is worth considerably more than 100,000 anonymous impressions from people scrolling past on a feed. The math on that trade is usually obvious once you've run both experiments.

4. POLICY WATCH

OpenAI and Anthropic Are Jointly Pressuring Australia to Let Them Train on Copyrighted Work

In submissions to an Australian parliamentary inquiry made public this week, OpenAI and Anthropic jointly argued for a limited copyright exception allowing AI training on Australian copyrighted material without permission. Both companies pointed to substantial Australian infrastructure investments, data centers and cloud facilities, while making their case. Australia has so far maintained that existing copyright law applies to AI companies the same way it applies to everyone else.

The leverage play deserves to be named plainly. We will build things in your country if you give us the copyright flexibility we're asking for. That's a negotiating position, not an argument about what copyright law should mean.

The substantive disagreement underneath this, and underneath virtually every AI licensing negotiation happening simultaneously in the US, UK, EU, and elsewhere, is the same question every time. Rights holders say training is a licensable use and AI companies should pay for it. AI companies say training needs some copyright flexibility or the technology becomes prohibitively difficult to build in any jurisdiction that doesn't grant it.

The US Department of Justice made essentially the same argument as OpenAI and Anthropic are making in Australia, filing a brief earlier this month that AI training on copyrighted material can qualify as fair use. That brief doesn't bind Australian courts or parliament, but it signals where American policy is leaning and that has weight in international negotiations.

Australia's parliamentary committee reports in November. The position they take will carry influence beyond Australia's borders.

5. PLATFORM STRATEGY

Spotify Is Testing a 72-Hour Premium Window Around New Releases

Spotify is testing Premium Early Access in India, allowing selected releases to be heard by paying subscribers for up to 72 hours before they become available on the free tier. Participation is not automatic: artists and labels collaborate with Spotify to choose which releases receive the window. The licensing requirement originated with label negotiations rather than Spotify's own strategic initiative.

Step back and appreciate how significant this is as a philosophical shift.

Streaming largely killed windowing as a music business practice. The model that replaced it was everything, everywhere, simultaneously at midnight. The differentiation between free and paid Spotify has mostly involved ads, audio quality, and control rather than which music you can hear and when.

Spotify is now quietly testing whether Premium subscribers should get the new record Friday while free listeners wait until Monday.

For developing artists, the calculation is usually straightforward. Maximum release-day reach across all tiers, free and paid, matters more than extracting subscription value from existing demand when you're still building an audience. A three-day exclusive doesn't help you find new listeners who don't yet know you exist.

For established artists with highly motivated fan bases, the calculation inverts. If your fans would pay for Premium specifically to hear your record three days earlier, that's a real monetization mechanism for existing demand rather than a discovery mechanism for new audience.

Watch whether this expands beyond India. If Spotify's data shows that a 72-hour window materially increases Premium conversions without hurting overall consumption numbers, labels will push for it in every market. The first crack in streaming's everything-simultaneously model may have just appeared.

6. ARTIST ALERT

Suno Confirms That What You Created Inside the Platform Helped Train V6

In its response to Universal and Sony's latest copyright lawsuit, Suno confirmed that V6 was trained partly on creations generated by users inside the platform, alongside licensed content from Warner, BMG, and Believe, and user preference signals.

This is new and materially different from what Suno said at the V6 launch.

When V6 launched, Suno described its user data contribution as preference information, the signals users generate by indicating whether they prefer one song over another. That answer left one category conspicuously unresolved: the songs that users actually generate inside Suno itself. Now we have the answer. Those creations, the output of users' sessions on the platform, were part of the training mixture for the next model.

Suno hasn't disclosed which creations were included, what share of the training corpus they represent, or how they were selected. The terms of service grant Suno broad rights to use submitted and generated content for purposes including improving its models.

The distinction for anyone using the platform commercially is worth stating clearly. Paying for a commercial license tier may establish your rights over a specific output. It does not necessarily mean your interaction with Suno concludes when you download the file.

The larger lesson applies beyond Suno to every generative AI tool. Before feeding original lyrics, demos, stems, voices, or unreleased masters into any generative system, the question isn't only what rights you have over what comes out. It's also what rights you're granting over what goes in and what gets created along the way. Read the terms before the session, not after.

7. PLATFORM TOOLS

GRAI Just Acquired Hangout. Streaming's Next Phase Is Participation, Not Access.

AI music company GRAI has acquired Hangout, the social listening platform where users enter virtual rooms and take turns DJing for one another. Hangout launched with more than 100 million licensed tracks from Universal, Sony, Warner, and Merlin. GRAI plans to layer licensed AI participation on top of that experience, allowing users to interact with and alter existing recordings while rights holders control participation and receive payment.

Place that beside the other developments pointing in the same direction this week and over the past month.

Spotify is testing Premium Early Access, asking whether the experience of getting music first has value beyond just having access. A Vinyl Bar in Shibuya raised $5.5 million so listeners can take recordings apart and reassemble them. Spotify is expanding video. GRAI wants social listening plus the ability to legally modify the song.

The basic streaming product, search, press play, receive recommendation, repeat, has been the dominant music consumption model for 15 years. The catalog is no longer a competitive advantage because everybody has it. The next competitive question is what the listener can actually do with that catalog beyond pressing play.

The platforms experimenting with participation, creation, social context, and interaction around recorded music are pointing toward where the next decade of music consumption is heading. Passive streaming is not going away. But it may stop being the only thing.

8. INDUSTRY STRATEGY

Sony Joined Disney, the BBC, and The New York Times to Help Write AI's Rules

Sony Music became the first music company to join ARIAM, the Alliance for Responsible Innovation in the Arts and Media, a coalition that already includes Disney, the BBC, The New York Times, the Financial Times, Condé Nast, Fox, and Adobe. The coalition's position is that AI innovation is acceptable but not on the premise that copyright stops applying when the customer is a machine.

The timing reveals something important about where major label AI strategy has actually landed.

Sony and UMG just filed another copyright infringement lawsuit against Suno. Warner, BMG, and Believe have active licensing deals with Suno. The same industry is simultaneously suing an AI company and licensing to it, depending on which major you're asking and which deal you're examining.

That's not contradiction. That's a two-track strategy, and it's actually sophisticated.

License when the economics are right, the protections are meaningful, and the deal gives artists real participation. Litigate and legislate when the economics are opaque, the protections are inadequate, and artists aren't getting compensated for uses of their work.

For artists navigating their own agreements with labels and AI companies right now, the same dual-track thinking applies. Engagement with AI platforms and tools isn't inherently wrong. Engagement without explicit answers to the compensation, consent, and control questions is.

9. PLATFORM TOOLS

YouTube Will Now Let You A/B Test the Actual Video, Not Just the Thumbnail

YouTube announced more than 30 creator tools at Made on YouTube this week, but one stands out as immediately actionable for anyone making video content for artists or for their own brand.

Creators will be able to test up to three different cuts of the same video and let actual audience behavior determine the winner. YouTube says more than 40 million title and thumbnail A/B tests have already been run on the platform. This extends that same data-driven optimization to the video content itself.

The workflow this unlocks is genuinely useful. Instead of making an educated guess about the best opening hook for a piece of content, you record one script with three different first-three-second approaches and let the audience settle the argument with their actual behavior. Same material. Different front door. Data instead of opinion about what connects.

YouTube is also adding AI feedback on unpublished videos before they go live, channel-matched title and thumbnail suggestions, a research feed showing what's currently working on the platform, and analytics designed to explain why content performed rather than simply reporting the numbers.

The platform is simultaneously expanding its Shopping affiliate program to 35 countries, with AI automatically tagging relevant products inside videos. The direction is consistent across every announcement: YouTube wants to be where content is created, optimized, tested, and monetized, without the creator or the viewer needing to go anywhere else.

For artist video strategy, the multi-cut testing is the feature to prioritize. Use it to turn hook selection from creative opinion into audience-verified data.

WHERE WE READ

Nine Signals is built from a curated stack of sources we trust to surface what actually matters in the music business. Here's where this week's edition came from:

Los Angeles Times — The artist AI compensation investigation that anchors tile 1 this week came through the Times' entertainment business reporting. Essential long-form journalism doing the work that trade publications don't always have the resources to pursue.

Music Business Worldwide — The GRAI/Hangout acquisition, Sony/ARIAM announcement, and Suno V6 training disclosure all came through MBW this week. The gold standard for music industry business reporting. Required daily reading.

Reuters — The OpenAI/Anthropic Australia submissions, Spotify Premium Early Access, and YouTube creator commerce expansion all came through Reuters. Essential for the policy and platform signals that eventually reshape music industry economics.

Korea Times — The Melon chart story came through Korean music industry reporting. A reminder that some of the most important consumer behavior data is happening in markets English-language trade publications cover inconsistently.

A curated daily briefing — The strategic framing connecting this week's AI rights, platform evolution, and participation stories into a coherent picture came through a personalized daily intelligence briefing that shapes our reading throughout the week.

Audacy/Acast/Annalect — The podcast ROI research came through the Audacy report citing Acast and Annalect econometric data. Apply appropriate skepticism to vendor-sponsored research while taking the underlying behavioral argument seriously.

We read widely so you don't have to. If you're building your own source stack, start here.

9 Signals drops every Friday via lpconsultingllc.music. If someone forwarded this to you, subscribe so you don't miss next week.