Can AEO Help Trigger Music Discovery for New Artists?

Understanding how to trigger discovery is as valuable to musicians as building a fanbase. And… I think my black belt in AEO would actually translate pretty cleanly to recommendation engines on music streaming platforms.

---

What Discovery Looked Like. Historically

Back in the day, albums dropped on Tuesdays – so Tuesdays were a big deal. I loved reading Billboard and R&R to see what was coming out, and it wasn’t unusual for me to head to my local record store that same week to pick up a new release. 

As a rock fan back in the 80s, I also got wind of new music via lyrics sheets printed in rock magazines. Side note: I woefully miscalculated the vibe of the G n R song I Used to Love Her when I read the lyrics. 

And then there was the Top 8 at 8 on FM 98 in Detroit. That list was mostly proven hitters, with the occasional new song landing on the list with a bullet. Or whatever OG deejays said. My point is music discovery usually came through known channels – conversations, magazines, radio deejays. But they were typically architected by industry insiders crafting trends, not personal tastes and preferences.

Today, Recommendation Engines Drive Music Discovery

Music discovery doesn’t work that way. Not anymore.

While I do find out about new music through links from friends and the praise team at church,  most of the time, my next favorite song – the song that makes me lose my breath and stop being able to focus – is gonna find me via the recommendation engine of the platform I’m listening to. I’ve come to expect it. I curate my engagement on streaming platforms to ensure it.

Back in my Spotify days, the platform was so good at matching music to my tastes that I’d spend Friday mornings just listening to the New Music Playlist and thumbs-upping anything that hit. Anything that I wanted more of.  Now, YouTube music is my go-to.

Preamble’s over. Let’s move on to the point.

River Crombie and the Sync License Rabbit Hole

Over the weekend, I was thinking about music – my music, the music of my musician friends, and how critical streaming platforms are for discovering music. Got me thinking about a song that’s been stuck in my head for months. River Crombie’s version of “Simple Man,” which is the song you hear in the title sequence of the Amazon series, The Terminal List: Dark Wolf. Love that show. Like, LOVE it.

The day before yesterday, River Crombie showed up in my YouTube feed while I was watching neuroscience videos. So I clicked. Then I clicked another one. Then another one. Nirvana’s “Heart-Shaped Box“. The Cranberries’ “Zombie“. Radiohead “Creep.“

A bunch of these emotionally heavy songs stripped down to an acoustic guitar and one guy sitting in a corner somewhere singing in a way I can feel it. Now, I am a sucker for boys with guitars. Always have been. Roots music.  Americana.  Folk. Soul. That whole world. My grandfather was a Blues songwriter – never saw him without his guitar. So, it’s in my bones. But as I’m looking at River Crombie – a new boy with guitar – I’m noticing that almost everything I’m seeing by him was uploaded in the last one to two years. 

And my first thought is: Who is this guy? Did he just come out of nowhere?

Then my brain immediately jumps to licensing. 

How did his version of “Simple Man” end up as the opening song?

Why not just use Lynyrd Skynyrd?

What happened there?

And now I’m down a completely different rabbit hole trying to understand sync licensing, publishing, and music placement.

The Metadata Problem

The funny thing is that I wasn’t actually interested in licensing. Not really. It’s a hot topic with indie musicians, and River Crombie’s a good example of this – didn’t know the guy before I heard his voice singing on my favorite show. Now I’m ready to buy him coffees on Patreon. 

But an idea stood up in my mind as I was thinking through his Americana roots. And that was this: how did the music director (or whatever that person is called) find River’s song? If it wasn’t a personal relationship, could it have been an algorithm? And if so, what data had to be tied to that song in order for machines to surface it? 

Because when I look at a lot of musicians I know, they don’t really think about metadata. They think about songs. The song is done. The album cover is nice. Let’s make a video. Let’s upload it. Move on.

But then, I compare the information available in the Song Credits tab for a local Detroit artist to what I typically see documented in the Song Credits tab for bigger indie artists, or even major artists. And there’s a huge friggin gap.

The writer credits are suspiciously thin. The band is listed as the artist, composer, and lyricist. No producer and studio info is listed. The artist bio reads like a book report instead of a marketing asset. And the biggest offender to me – you can’t find the lyrics anywhere – you have to try and write them down in a notebook like it’s the 90s.

That’s when I saw the gap that needed to be filled. And in that gap, just enough space to build the bridge that connects my current world or SEO and AEO to that world. 

Discovery Versus Visibility

I took the conversation to ChatGPT. Tell me what information is needed, Maverick. What do distributors usually ask for that you need to have on hand? Maverick gave me a thorough breakdown of the info needed – information that audiences would need to have access to. Information that would ensure songs get played.

But something in the framing was bugging me.

The entire conversation was built around the assumption that the band had an audience to play music for. The assumption that they’d already built a fan base and was releasing music to that audience. 

But what happens when they haven’t built a fanbase? We’re not judging whether they should build one – every brand needs an audience. But the reality is that many musicians don’t build a large following before they release music. They release the music and hope folks will stream it when they find it later. 

So, my mind shifted – assuming the musician has no audience, how do we get the machine to surface their music to the right audience so that they can start building one?

If we treat YouTube Music’s never-ending feed like the social media algorithm it is, how does one need to tag the music – or content, as I think of it – so that fanless musicians have songs that still get surfaced in data-driven recommendations? Because if I can figure that out, I can solve machine problems for musicians, too.

That’s when my mind shifted from visibility to triggering discovery. Making music recommendable when no one is looking for it. That’s not a people problem – that’s a math problem. And Mama loves math problems. These kinds, at least.

Rethinking Distribution As a Discovery Strategy 

At that point I shifted my thinking around distribution from access/visibility to discovery. Visibility is getting seen. Discovery is getting matched. 

A song can be sitting on every major streaming platform and still be functionally invisible because the system does not know what to do with it. That’s the part that has my attention. 

I’m trying to understand how recommendation systems inside music streaming platforms decide who should hear a song first.

Musicians know to tell people how to get the song on Spotify. Get the song on Apple Music. Get the song on YouTube Music. Get the song on TikTok. Get the song everywhere people listen to music.

And yes. Fine. Great. Necessary. But being available is not the same as being discoverable.

What if nobody knows who you are? What if you don’t have an audience waiting for the release? The only path to having your music heard and appreciated is making it findable. In a situation like that, the metadata has to do a lot of the heavy lifting, helping the machine decide:

Who does this sound like?

Who might like this?

Where should it go?

And if that’s true, then the quality of the metadata matters a lot more than most musicians probably realize.

A Million Genres I Don’t Know

That led me into another problem. There are entire genres that didn’t exist when I was active in music. Or maybe they existed and I just wasn’t paying attention.

• Amapiano.
• Bedroom pop.
• Alternative R&B.
• Neo-folk variations.
• Indie subgenres.

All kinds of things. Things I know exist. But things I couldn’t pick out of a lineup. 

The first time I even paid attention to Amapiano was when I was trying to figure out how to categorize Willow Smith’s Empathogen. One question led to another. One genre led to ten more. One artist led to twenty more.

And now I have this growing realization that if I’m going to help musicians with publishing administration and distribution (i.e. teaching machines about musicians), I need to get much better at understanding genres I don’t personally listen to.

Genre is metadata. It’s classification and discovery infrastructure – taxonomy. And if I don’t nail primary genre and secondary genre — if I go too broad, or too old, or too lazy — then the song may get sent to the wrong shelf.

That matters because listeners don’t only discover music by genre anymore. They discover music by vibe, by mood, by use case, by memory, by identity, by emotional state, by proximity to something they already love.

I have to know those signals and classifications like the back of my hand so I can teach the machines where a song belongs before enough people have listened to it to generate behavioral data.

And for a new artist with no audience, that early classification may be the difference between getting tested with the right listeners or disappearing into the catalog.

The Secondary Genre Question

What I’m particularly interested in is the secondary genre. Most of the musicians I know are older musicians. Their work tends to originate from traditional genres.

R&B.
Soul.
Gospel.
Blues.
Country.
Rock.

But modern music rarely stays inside one lane.

The first time my daughter played Kane Brown for me, I remember immediately saying, “That’s not country. That’s a Timbaland beat.”

She thought that was hilarious. “You’re such a Boomer, Mom.” I’m not. But I was serious. I could hear the old school syncopated rhythms and unexpected pauses in the percussion track – a style made famous in the early 2000s by Timbaland – laced all up and through Kane Brown’s “country” song. 

The same thing happens when I listen to Chris Stapleton. Tell me “White Horse” isn’t giving Richie Sambora vibes and I’ll smack your idiot face. Country music getting an 80s arena rock makeover is making me like country music. 

But then, there’s the subgenre my friend contemptuously calls “face-tat trap country” which…

[sigh]

Okay. Whatever.

But I’d need to know that language – trap country – in order for me to help machines identify a piece of music as a trap country song. So it doesn’t end up in the feed of a sixty-three-year-old George Strait fan. 

So now I’m asking a different question: How do recommendation engines think about those overlaps?

A country song may pull from gospel. A soul song may carry folk structure. A worship song may have pop production. A blues record may feel like classic rock.

An Americana artist may belong beside an acoustic R&B artist more than beside someone who wears the same genre label.

So when a musician picks one genre from the drop-down menu and hurries through the rest of the fields when pushing their song out with a distribution service, are they accidentally flattening the map? Are they telling the machine less than the machine needs to know?

That’s where the secondary genre starts to feel important to me. It refines the routing. Because the primary genre may tell the system what world you come from. But the secondary genre may tell the system who else might love you.

Looking for Metadata in Spotify

At some point, I decided to stop theorizing and just go look. I opened Spotify. Stephen Day was already on my screen, so I clicked through.  I hoped to find a place where artists could describe individual songs. Some kind of field where you could say, “This song sits somewhere between southern soul, acoustic pop, and 70s singer-songwriter sadness.”

I didn’t find that.

What I found instead was his bio. And Stephe Day’s bio was fascinating.

Different sections of his bio talked about different things. Growing up in church, his father being a pastor. Southern roots. Then there was an entire section dedicated to influences. Then another section talking about artists he had toured with. 

And I remember staring at it and thinking, “Holy crap! This is adjacency architecture.”

The bio wasn’t just a biography. It was metadata written in paragraph form.

• Southern church roots.
• Influences.
• Touring partners.
• Collaborators.
• Comparable artists.
• Audience overlap.

Everything I hoped would exist at the song level was being established at the artist level.

The bio was helping the system understand where Stephen Day belonged. Who he sounded like. Who his audience might overlap with. Who he should stand next to on the shelf.

Important friggin find – glad I looked.

And another aside – doesn’t it look like the algorithm is dictating tours? Brand and Monica? Huge audience overlap – nostalgia 101. Chris Stapleton and Allen Stone – soulful songwriters who play guitar. Just saying – live shows out here lookin’ like my YouTube Music quarterly recap.

If you know anything about me, you know I’m a stickler for a founder having a good canonical bio that helps machines understand you.

Unlike companies where you’re trying to find your blue puddle and stand out, musicians are trying to help people and machines understand who they are most like, not how they are most different. Because music is a vibe, and the goal of the algorithm on a streaming platform isn’t to find the unicorn, it’s to match audiences to songs that are least likely to kill their vibe. 

In a machine-mediated discovery environment, a musician’s bio is where context and adjacency get built. It’s where they name the lineage and surface the sonic neighborhood. It’s where they create recognizable adjacencies for humans and machines at the same time.

That is architecture.

How did I not see that before?

The Problem Is Still Recognition, Though

The cool thing about music is that each song gets its own set of data points. The song itself becomes an entity. (Talked about this a bit in that AC/DC licensing blurb in this article on the conditions needed to become recommendable to machines.) Which means, a song can be surfaced to listeners who may not receive recommendations for other songs in a musician’s catalog.  

The Entity architecture of the song is as important as the architecture of the artist. But at the end of the day, the goal is still the same: recognition, machine legibility, and helping systems understand what they’re looking at.

I started the day thinking about music distribution. I thought I was learning about publishing administration. I thought I was figuring out how to help local musicians get more organized in order to find more monetization opportunities.

And I am.

But underneath all of that is the same problem I’ve been obsessed with for years: Machines don’t recommend what they don’t recognize. 

Whether we’re talking about a person, a company, a city, or a song, the challenge is remarkably similar. You have to teach the system what the thing is. You have to teach it what it’s adjacent to. You have to teach it where it belongs.

And that means someone like me – who has spent the last 15 years in SEO and the last two in AEO, who reads music as content to optimize – can prove valuable to musicians.

As unsexy and unmusical as that sounds, it’s becoming more and more important for independent creatives.