MBW Views is a series of op-eds from eminent music industry people… with something to say. The following MBW op/ed comes from Sammy Andrews, CEO of Deviate Digital and a long-standing MBW columnist. Below, she argues that the real, unspoken AI battle in music may be less about ownership than about the difference between distribution and desire…
The music business has spent the past few years publicly arguing about AI and ownership. Who trained the model? Who owns the output? Who gets paid? Those questions really do matter. But they’re not the only ones.
AI is forcing the industry to confront something it has been quietly avoiding: are streams a measure of audience desire, or just a measure of distribution?
For much of the music business, streaming data has become a proxy for audience preference. For many consumers, it has become a proxy for taste itself. But if recommendation systems shape what people discover before they’ve consciously formed a preference, the data isn’t measuring fan sentiment anymore; it’s measuring the platform’s own output, reflected back at itself. The industry is reading its own reflection and calling it the audience.
Demand has always been shapeable. The difference now is scale and intimacy. Previous gatekeepers influenced culture collectively. AI is doing it individually, continuously, and at planetary scale – personalizing influence for millions simultaneously, learning from every interaction, adapting in real time. That is a very different kind of gatekeeper.
The shift from “search and click” to “ask and receive” is well underway. Spotify is integrated into ChatGPT. Google‘s AI Overviews are materially reducing click-through rates. When a fan asks an AI assistant what to listen to, they get an answer, not a menu.
Researching this piece, I asked three popular AIs to recommend music for the evening. They gave confident, thoughtful suggestions. When I asked why, the answers became much more vague.
One told me it was drawing from “a mix of cognitive psychology, neuroscience studies on productivity, and collective human behavior”. Another described access to sonic metadata – BPM, key, energy scores, danceability – as though platform data were independent musical insight.
They knew their understanding had been shaped by journalism, criticism, and fan communities, but couldn’t trace any individual recommendation back to a source. The idea of one artist being “essential” or another “transcendent” had been absorbed from somewhere, blended with everything else, and re-emerged as something that felt like taste. That’s fascinating. It’s also slightly unsettling.
Recommendation has always carried authority. AI introduces a new kind, one that confidently shapes discovery while being unable to fully explain itself. The recommendations of critics, fan communities, and editorial voices don’t disappear inside these systems; they get absorbed, blended, and redistributed at machine scale.
The authority remains, even if the attribution doesn’t. Journalism has become training data for taste formation rather than a destination in its own right.
If the primary discovery interface moves above the DSP into a conversational layer, the new bottleneck isn’t the playlist or the algorithmic push, it’s the prompt response.
Chart positions have been materially shaped by playlist placement for years. Algorithmic promotion can create outcomes that look entirely demand-led, even when recommendation systems did much of the work. The data to distinguish lean-forward listening (direct searches, saved tracks, intentional plays) from lean-back listening (autoplay, radio mode, algorithmic feeds) has existed inside the DSPs for years. Its absence from charting methodology is a commercial and political choice, not a technical limitation.
The conversation about streaming fraud is important, and its scale is larger than most people in this industry are willing to say publicly. But fraud and algorithmic demand aren’t really separate problems.
“You don’t need to break the rules to manufacture demand. The system, operating entirely as intended, is already doing that.”
Bad actors learned to game the system because the system rewards the same signals either way. You don’t need to break the rules to manufacture demand. The system, operating entirely as intended, is already doing that.
The danger isn’t just one of measurement. AI optimizes for frictionless compliance, music that is perfectly pleasant and completely unchallenging. The weird, volatile artistic risks that actually move culture forward don’t score well on an engagement metric. That has consequences well beyond the charts.
None of this requires conspiracy or bad intent. It just requires companies optimizing for what they’ve always optimized for.
We have all spent years mastering how to market to algorithms as well as fans. The infrastructure of algorithmic demand isn’t coming, it’s here.
A stream generated through recommendation tells us something about reach. It tells us far less about desire – unless you’re actively reading the skip, save, and like rates that sit alongside it. For an industry that uses streaming data and chart performance to drive commercial decisions across every sector, that distinction matters more than most people realize.
Every AI-in-music headline is still about generation. Almost none are about what the same models are doing to the people on the other side of the speaker.
AI can already read a fan’s digital footprint – listening behavior, social activity, purchase history, location – and predict lifetime value, churn risk, and exactly which message they’re most likely to respond to.
The commercial temptation is obvious: individualized pricing for merch, tickets, VIP experiences, and artist interaction, calibrated to what a model thinks each fan will pay. Dynamic pricing and behavioral segmentation are already standard practice in travel and advertising as well as secondary ticketing. Is that going to be part of any superfan model?
Fans’ listening and purchase patterns are being aggregated into behavioral profiles of real commercial value, used to shape what they’re offered and potentially what they’re charged. Legislation like GDPR gives regulators some tools, but it was never built for this.
Then there’s voice cloning and conversational AI. Artists may soon offer an always-on relationship no human could sustain. AI companion apps have been downloaded more than 220 million times globally, with category growth of nearly 90% year-on-year.
Usage is particularly high among younger demographics. A polling survey from the Autonomy Institute found that four out of five young people in the UK have used an AI companion. Of those, roughly half describe themselves as regular users. Once that capability attaches to fandom, it won’t move slowly.
The consent frameworks don’t exist. The NO FAKES Act in the United States, for example, targets unauthorized deepfakes… not what happens when an artist or their estate actively deploys one.
You could argue that DSPs and emerging conversational AI platforms have every incentive to keep this debate focused on generation, licensing, and attribution. If the industry spends the next three years litigating training data and output ownership, the companies building recommendation systems and fan-data pipelines will have locked in structural advantages before anyone has seriously asked whether they should.
The practical implications are already arriving. Artists and labels wanting to be surfaced by AI discovery systems are looking properly at their metadata, bios, and knowledge base presence as Generative Engine Optimization (GEO) for music. MusicBrainz compliance, DDEX tagging, Spotify bios written for natural language processing rather than human readers. If that sounds like what SEO did to publishing fifteen years ago, that’s because it is.
I genuinely believe that if algorithms fully mediate discovery, then live music and genuine fan community become more valuable. An algorithm cannot replicate being in a room with other people when something real happens.
That’s not a sentimental argument, it’s a commercial one. The operators who position live music as a conscious counter to algorithmic culture, rather than simply the thing AI hasn’t replaced yet, will be ahead of those who don’t.
But the industry needs to be calling for transparency requirements for algorithmic recommendation; for fan data consent frameworks built for this moment, not the last one; for a demand-side working group with actual authority. None of this is radical. But the window to act before the infrastructure becomes too embedded to shift is not unlimited.
It’s time the music business discussed the demand-side impact of AI properly. What AI is doing to discovery, taste, identity, and the relationship between fans and platforms is structurally bigger, moving faster, and barely on the agenda. The intermediary isn’t simply deciding what’s available or visible. It is changing the preferences that drive demand itself.
The labels, managers, platforms, and artists that thrive in the next decade may not be the ones that adopt and adapt to AI the fastest. They may be the ones that retain the deepest understanding of human behavior alongside it all.
Because if AI shapes what people discover, what they listen to, what they talk about, and ultimately what they buy, the most important question facing this industry is no longer who owns the music. It’s who gets to decide what people want in the first place.
This article originally appeared in the latest issue of MBW’s premium print publication, Music Business Worldwide Magazine, which is out now.
Music Business Worldwide Magazine is available as part of a MBW+ subscription – details through here.
All MBW+ subscribers get digital access to our Music Business Worldwide magazine, with six issues released each year. Music Business Worldwide
