I Found Out Why Spotify Killed AI Music feat. Sam Duboff
Discover how Spotify’s new AI Personas feature is reshaping music discovery by distinguishing authentic human artists from AI-generated identities on the platform.
Quick summary
Spotify’s AI Personas initiative introduces a clear label for profiles that are AI-generated identities, addressing listener frustration with fake artist accounts flooding the platform. This move ensures that algorithmic and editorial support prioritize genuine human artists, preserving the integrity of music discovery. The feature combines manual artist disclosure with automated detection to identify AI personas, preventing deceptive profiles from gaining unfair exposure. The policy carefully separates the identity of the artist from the creative use of AI tools, encouraging transparency through AI credits when AI assists in music production. Spotify aims to foster a trustworthy environment where listeners can confidently connect with real artists, while still allowing space for creative AI experimentation under clear labeling.
Auto-transcript(English)
All right, so Sam, uh you guys just announced like what I think is the biggest like AI policy change it since AI music has been a thing. And And you guys rolled out this thing called AI personas where you are going to be allowing, I think as of today in Spotify artists um profiles can go in and manually say if they're an AI persona. And then starting in September, you're going to start also having like automatic detection either through human human flagging, but also like I'm guessing you have your own internal AI that flags AI. Um and then uh there's there's there's a lot of nuance to this when I when I read it. Um But but like the big the big kind of takeaway is to me this seemed like a really good way to solve the problem of companies that flood the market with like dozens or hundreds of fake personas, I guess you you guys would call them, where it's just it's a fake name, fake picture with like 50 music releases a year, and they have like a hundred of these profiles. Um those types of accounts now are no longer going to be eligible for algorithmic coverage or editorial coverage, which is huge. Like I uploaded a video this morning as a filming, and and the amount of like excitement was kind of unreal. Um but you know, just first like in your own words, like how how do you describe this feature, and then we'll get into like kind of the nuances of it. >> Um yeah, really really excited. We've been working on this one for uh a while. Uh as we talked about as on your show back in in March, and we touched on the complexity of of AI and music um a bit. It's a a complicated topic, and what we're trying to do is yeah, you know, those sorts of accounts you're talking about um are just the ones that were we're looking at. Um what we [clears throat] have heard consistently from listeners is they hate the feeling on streaming services where you see a song, you listen to it, maybe you like it, maybe you don't, but you kind of see an image, you think you're listening to a real artist, and then only later do you find out online or through a Reddit thread that that actually wasn't a real person. Um, and actually was a AI identity. Um, when it comes to how music is made using AI, there's a pretty broad spectrum, right? Everything from like mixing and mastering to generating or, you know, sometimes buying a looper sample or beat that might be AI generated to specific stems where you might use AI as a tool to transform a keyboard melody that you played into strings um, through to you know, getting help with vocals, uh, electronic artists having an AI top line through to, you know, full uh, entirely prompt generated music. There's a pretty wide spectrum when it comes to how artists can use AI as part of their toolkit creatively. We've launched AI credits a few months ago um, that speaks to that type of nuance where we're getting tens of thousands of AI credits every day now where artists can share the way that they're using AI. But, when it comes to identity, things are a little bit more clear cut. There isn't this spectrum in the same way. Of course, there's edge cases that people can debate, but either you are uh, real human artist building a career in music that I can talk to in real life or you're not. Um, and that's what today's all about. In April, we launched verified by Spotify, our new verification program where um, we have a team of human reviewers who are going through and they're making sure this is a legit human artist. The person pictured in the profile is the same per- person you can talk to in real life. They're often touring, building a career, uh, have a, you know, really deliberate and intentional artistic intent. That's that verified green badge that you see at the top of profiles. This is the other side of the coin, and we felt like after verified, this was still that gap, which is um when it's an AI-generated identity. And, you know, there probably are some artists who will find really cool creative use cases um for how they could use AI identities. You could picture a whole creative process where, you know, however you're making the music, you might want to display that identity um using kind of an AI identity or some kind of AI persona. That's okay. We just think listeners need to know that. And then, by default, our algorithms and our playlists are meant for human artists, authentic human artists. We're betting big on human artistry. So, AI personas, people uh could have that creative intent, go for it. If listeners want to go seek it out and actively listen to it, we think they should have the right to do that. But, our algorithms, our playlists are for legit human artists, and we want to make sure listeners never have that experience on Spotify where you see this photorealistic image, you think it's a real artist, you find out it's actually a fake AI identity later. Yeah. >> Yeah, totally. Uh and and there's kind of two two like, I guess, edge cases or or like unique scenarios uh because technically, you're not labeling AI music with this this this update. You're You're tagging people who are pretending to be someone else they're not using AI. Which is like 90% of the slot. So, like as a as a side effect, it is really like kind of solving a big part of the AI music problem. But, um the edge cases of that is like what what happens if you are uploading human-created music under a, you know, AI-generated uh image or or maybe not even AI-generated, maybe um just a graphics, right? It's It's an anonymous profile. And then, on the opposite side of that, what if you're uploading human-made music, or sorry, AI-generated music under your own name and picture or AI generated music where it's like a persona, like it's a band. So, you're not like hiding behind anything. You're not using AI's for the visuals. >> Yep. >> But, it's like a AI band and just has graphics that are not AI generated. So, those are kind of two extremes on the other side where um that's what me and you will I thought about. And then off my video like this morning I had a bunch of comments being like, "Well, what if I'm doing this? What if I'm doing this?" And then on the flip side, what if someone's doing like a 50/50 blend where some of these songs are human-made and some of them are running through Suno. And then the the the the public perception is like it's them, but like some of the visuals are AI generated, right? There's all these weird combos. So, like you got kind of preset uh defaults for how you would treat some of those cases? >> Yeah, there's uh certainly a million permutations. Um uh we're really trying to separate here the identity versus how the music's made. Um I think when it comes to AI generated identities for the most part right now, it's also heavily AI music. Often there's there's always a human there who's creating the the persona, but um you know, and maybe there's scales of how much kind of traditional music making they're using versus um AI tools, but the AI persona badge and label, that's for the identity. Um so, if you uh if you know, you decided to make a profile as Andrew and it's a you know, even if it's a stylized photo of yourself, right? If you um you know, have a certain outfit or visual identity you're going to use, but it's still you and then the song DNA and the credits you're crediting that full name so people know who that you are. In your bio, you're kind of sharing "Hi, I'm Andrew. Here's where I live. Here's where I'm from." If you are hopefully using AI credits if you're significantly using AI in your production process, um you know, I think listeners have some openness. They just want to know, but some people will be okay with knowing who the person is and when there's transparency about the tools, I think there's lots of artists who are experimenting with AI authentically in really cool ways. Um >> Yeah. >> The other side of the coin is when it's an AI identity, and no matter how you're making the music, listeners are telling us overwhelmingly they don't like the feeling when you look at a photorealistic image and you don't know if it's a real person or not. Um and so, you know, there's a million edge cases for sure. Um we didn't want to wait to figure out, you know, you could debate yourself to death and never ship anything. We think this is a a big improvement that listeners have been asking for that's going to help protect Spotify for for human artists. >> Yeah. >> In general, if it's an AI-generated identity, there's definitely space for you to release music on Spotify, to appear on Spotify, you know, for listeners to actively seek you out. We just think that should be clearly labeled so listeners always know. Um >> Yeah. >> And if you're a human authentically presented artist um as yourself, you might use a pseudonym, you might use a band name. That's why, you know, talking about AI personas today, it really isn't really a one-off. This is coming at the end of launching song DNA, performance videos, um new concert recommendation features, uh verification. All these things where collectively we would have all these ways to show your authenticity. Um and you know, the big bet we're making is that in the age of AI where across industries, there's just this like overabundance of content, people feel overwhelmed, they don't know what to real or what isn't. We're trying to place a big bet that what listeners are going to care about the most is knowing what's real, knowing who the human artists are. You're going to open Spotify, see an artist is verified, see the song DNA and find out the amazing producers and musicians that they work with, then you're going to see their tour in near you and you're going to go see them live, you're going to watch a performance video uploaded through Spotify for artists, you know, that's the the journey. So, I'm happy to go into any of those edge cases, too, but big picture, if it's a AI-generated identity that doesn't match the person behind the music in real life, that's an AI persona. >> Yeah. >> Um and for artists who are experimenting responsibly and transparently with AI and authentically presenting themselves, love it, go for it. Um if you're sincerely using AI, you should use AI credits to share that with listeners. >> Around maybe March, April, there was a huge, like, purge in the in the Spotify algorithm playlist for um AI-generated artists that I noticed, like, across like dozens of different accounts that all of a sudden, like, in one case, 300,000 streams a day from radio, and within weeks down to 5,000. Um so, pretty, like, very dramatic, 97% decrease. And and then again, um around July, early July, there was another kind of reckoning where a lot of other things got got the got slashed. Um and you guys never even announced that. That was just something where where all the all the people, you know, cuz I have an agency, we we will promote anything if we think it'll it'll work. So, like, we see all these AI artists and labels and stuff, and um they're like, "What's happening?" I was like, "I don't know, but Spotify hasn't like announced anything. They must be doing something with how they're handling AI music." Um and, you know, I thought it was interesting cuz you guys probably would have got a lot of brownie points if you if you posted online that you you adjusted the algorithm to uh be more in favor of human artists, but that seems like what happened. >> Yeah. >> I mean, we're we're never looking for for brownie points. We're trying to build the best experience for human artists and for listeners. And so, you know, what we did announce last fall, I forgot if it was September or October, was um a whole kind of set of new protections for artists where we're trying to fight spam, slop, impersonation, and deception. Um and throughout the last year, we've made a number of changes in our algorithms over time, all with that in mind. Um how do we elevate and make sure that there's more recommendations for authentic human artists? And how do we remove from recommendations content that listeners tell us feels like low-effort content, feels like spam, feels like slop. And certainly anything that's impersonating we're removing from the platform. Um you know, uh over the last year alone we removed over 75 million spammy tracks from Spotify altogether where we saw egregious uh kind of spam tactics. Um we absolutely um believe artists, songwriters, producers should be in control of the choice of if and how they use AI in their music. Tons of the biggest songs on streaming have heavy use of AI. Um right? A lot of the biggest artists in the world are using AI. And studios all over the the country, certainly not everyone, but there's lots of artists who are using AI as a one tool in their tool kit in a creative process. That's great. Um we always support artists getting to experiment with the technology they want to. What we don't want is low-effort content. What we don't want is content that feels deceptive to listeners, right? If there's like a heavy use of AI or if an AI identity isn't real and you're not disclosing it, if you're not using our AI credits system, those are the sorts of signals that might tell us, oh this is low-effort, this is the sort of recommendation that um listeners might feel like um isn't the experience they want out of Spotify. So, different signals might, you know, as we add or remove them from the algorithms might impact different artists differently, but you know, if I was an artist heavily using AI in my creative process, um I you know, definitely they should feel like there's uh a role for them on Spotify, but they should be doing it transparently and authentically. They should um either have an authentic human uh um identity, show who they are, even if they have a cool, you know, creative project around it, use AI credits so listeners know. Um and you know, you can authentically find some fans based on that premise of trust if you want to use an AI identity. There's still opportunity on Spotify if listeners seek you out actively, um but you know, right now from what we're hearing from listeners, we don't think there's a role for that type of music in programming unless the listener has gone and followed those artists and told us they really want them. Um so, uh you know, that's the the sort of um platform we want to build where listeners can trust what they're listening to. AI is going to play an increasing role in the creative process on that spectrum of use cases, um whether it's just, you know, a plugin you're using in your DAW, uh you know, uh an example I see a lot online is someone buys a loop uh or sample on a marketplace, they didn't even know it was AI-generated, and then some system flags them as AI. Um if you think of like, you know, pitch correction technology, um uh I'm really interested in the idea of, you know, some artists are doing language translation where you can sing your songs in a different language that's using AI. Be transparent about it. Be authentic about it. Be clear on Spotify and all streaming services what your process is, and then let listeners choose. We want fandom on Spotify to be based on authenticity. When listeners have that information, they can make the choices. >> I like how you one one of the kind of nuances of this rollout is that when you guys start automatically detecting and and labeling um and the human reviewing and all that, um there's there's like a thing in the I think the song DNA part, whatever, that shows if they self-disclosed or if they were automatically tagged by Spotify. So, it kind of the people who try to kind of pull a fast one are kind of going to be labeled as non non-transparent as a result. >> Yeah, so on the artist profile, it's going to say AI persona at the top. When a listener taps it, they're going to say the artist has shared their AI persona and I hope people give them some credit for doing that transparently. Um and if they don't, it's going to say because we aren't sure, it's a likely AI persona. And um to start, we're just going to be reviewing profiles that have a more material audience because then we can focus our review efforts and make sure almost all the personas that people are actually seeing are labeled. I hope listeners do pay attention to Okay, if I'm thinking about listening to an AI persona cuz I happen to be the sort of listener who's open to that. I think they will be more open to artists who are sharing that openly rather than those that Spotify has to go in uh and apply the label. I will say, you know, we've launched AI credits a number of months ago. We launched it and candidly, I thought it was the right thing to have the opportunity for artists responsibly using AI to share it. I had no idea. You hear online that no one would ever share that they're using AI. We kind of like had no idea what we'd see. And it's like truly shocking the level of self-disclosure that who have access to AI credits are doing. >> That's what I would I would have expected, too, that almost no one would would flag it, but you're saying that like Do you have any like I don't know, any any benchmark on like how many how many people have tagged it, how many songs have tagged it, or what percentage? I don't know if you have any kind of off-the-cuff numbers. >> Yeah, we're getting, you know, several tens of thousands of AI credits every day. Um it's a solid percentage for participating distributors. There's only a few distributors that support it right now. Um 99.9% of the AI credits are on songs that also have human credits. Um so, almost never are you seeing AI credits and that's it. There's the human credits and then they're sharing the role. Um and we're seeing everything. We're seeing AI guitars, AI drums, AI vocals, AI synth, uh AI songwriting. Often they'll be in combination with a human, so maybe the human songwriter is sharing the part of the songwriting process they got help from AI. >> Yeah. >> And I think there are, uh, you know, big set of artists using AI as one part of their creative process, right? We're not talking here about entirely prompt-generated music. We're not talking about industrial production AI where you're mass exporting music and flooding streaming services with it. Um, AI credits are being used by artists who want to build an authentic audience. You know, I think it will make sense in the long term that artists heavily using AI are going to want to embrace these features cuz if you >> Yeah. >> aren't sharing how you're using AI, when a manager, when a label comes to sign you, they're going to wonder about your process, right? If you >> Yeah. >> don't share where, you know, the, you know, if there isn't a certain instrumentalist, but you can hear it in the music, people will ask that question. If you can't sing the song live and don't share it's AI vocals, then, um, you know, at some point the gig's going to be up. Uh, and so, we're trying to set up a system, uh, we're proud to be leading the industry on these sorts of self-disclosure, um, fields where artists can be upfront, listeners can make decisions, um, and, you know, if you're an authentic human artist doing things the right way, responsibly, and you're using AI transparently, um, you know, obviously listeners have different feelings right now in the world about AI. We feel like our role is to enable that sort of transparency and then let listeners decide. >> Yeah. Uh, one one kind of a wild thing is this week there was there was a lot of like AI updates, right? There's there's this roll out from you, but then recently Suno said they were going to start watermarking their releases and then sharing those watermarks with other companies to make it easier for distributors and DSPs, etc., to to flag. Uh, and then they they announced they were going to do download limits for their plans. And and I was expecting the download limits to be bigger than they are, but they they actually went pretty pretty limiting with it. Like for Suno, you can get to either 20 downloads a month or 60 downloads a month on different plans. And so, um that's pretty severely cuts back on the the amount of slots someone can either create. Um they have to be a little more intentional with what they're they're downloading. Um but it it felt like the timing was kind of wild on this cuz so, Suno had that massive update and then a couple days later you guys announced this update. >> Yeah. >> And um I I heard some people were saying like this is kind of all based around some EU law that passed around like um AI disclosures or AI transparency. And I don't know was that a factor here at all in terms of the timing of this or >> It definitely wasn't a a factor for us. I think AI news is happening every week in the music industry now. So, whenever we announce that, there'd be something else that week. Um uh you know, I haven't seen um what the Suno watermarking specs going to be. Looked like from their language it's going to focus not necessarily on all Suno creations, but things that would help with kind of trust and safety issues like infringement type cases or um uh ripped content, that sort of thing. So, we'll have to see what what that ends up being. I definitely applaud we definitely applaud the download limits. Um I think they had a phrase in their blog post I liked that was around you know, trying to reward intentional music creation, right? I think that's what the industry is centralizing around. If it's intentional music creation, human led, artistic intent, AI's one tool of many that you're using. Um I think [clears throat] that's pretty reasonable. I you know, don't see a reason why anyone needs to be able to download hundreds of tracks uh that are AI generated. So, definitely applaud them for that. Uh and yeah, regulation's always evolving. I think that piece of regulation talks about services that are making AI music rather than platforms, but um I think you know, regulation from around the world is a a good thing cuz it's that sphere rules for everyone, makes it clear what the rules of the road are. Um for this particular um launch, you know, I think I even teased it a little bit back in March on your show. We've been working on it for a while. AI identity and trust, you know, in January we published our top priorities for the year. That was one of the top ones and you've seen us do artist profile protection, song DNA, verified by Spotify, AI credits, now AI personas. This is the next in a long line of um investments we're making in the space. >> Awesome, man. Well, well, thanks for coming on and clearing this up. Do you have anything you want to leave the people with? >> Um oh gosh. Uh always happy to be on the show. Thanks for getting out the word. Yeah, I'd say like, you know, uh I hope if you're an artist listening to this and you um are building your career the right way, you're playing by the rules, you're um showing your creative process transparently, whether that's using AI tools, no AI tools, whatever you want. Um I hope you feel like Spotify is the platform that's investing in you. We want our algorithms and programming to to benefit you. We want authentic artists to um uh really feel Spotify is the place that's supporting them. So, um uh and I hope if you're a bad actor watching this to try to figure out how to game the system, I hope you go somewhere else. Uh we're we're trying to make it as hard as we can for you.
Grow your music with a proven ad system
Learn the Meta ads and streaming growth playbooks Andrew uses across artists, labels, and agency campaigns.
Keep exploring
