Brought to you by ten years of playlists, a collection of saved Instagram Bangers and an incredibly poor sleep cycle.
Poking around streaming service data to learn how I discover and consume music.
As much as I may hate to admit it, I spend somewhere between two and three hours of my day on Instagram. While most of it is some combination of the requisite doomscrolling, NBA highlights, and general shitposting, it's also become a surface area to stumble across new music.
And what do you do when you come across a great social media post that you can't fully consume appreciate in that moment? You save it for later; because memory is a fickle thing and searching for one specific post is a pain and there's no real downside to saving, is there? I mean it's all digital space anyway and this way it'll always be handy if you need to return to it, so what if you save like 3 posts an hour.
(Okay, that may just be me. Anyway.)
I saved my first "Instagram Banger" a little more than two years ago; a cover of Blondie’s “Call Me” by The Last Dinner Party on triple j's Like A Version radio show. Since then, I've saved 382 more posts to that collection, and 55 of the songs in them are now in my Spotify library.
Honestly? Sheer curiosity. That social media algorithms impact how artists produce and promote their music is not new information. Still, there does exist a certain ... snobbery around "reel music" and with the amount of discourse centred on human taste in the last few months, I wanted to see how that impacted my own choice of tracks.
The quickest way to do this was to compile a playlist with my Instagram saves, and compare it with my Liked Songs on Spotify. If the same song featured on both, it counted:
A couple of trends surfaced almost immediately, courtesy of Claude:
41 of those 119 songs had made it to my Spotify, about a third. Awesome. Job done.
It's at this point that I really should have gone to sleep.
However...
This exercise couldn't really be called attribution, simply because a bunch of those songs already existed in my regular rotation before the post went live or was saved. Or the more intuitive possibility that I was simply saving videos of the music I already liked. Comparing two playlists in isolation wasn't going to cut it.
I needed to find more data.
A couple of Google searches later, I came across Exportify, which let me pull all 216 of my saved playlists into individual CSV files with specific timestamps for when each song was added plus a whole lot more columns that looked very intriguing:
Track URI │ spotify:track:2QOTLNlbpsTJq074uw3yLY │ spotify:track:6dWilYAxP2aJbTTt0UsatK ISRC │ USC4R2665592 │ QM6N22553527 Track Name │ Blackberry Marmalade │ Lover Girl Album Name │ Cry Baby │ Lover Girl Artist Name(s) │ Vince Staples │ Laufey Release Date │ 2026-06-05 │ 2025-06-25 Duration (ms) │ 232933 │ 164400 Popularity │ 60 │ 76 Explicit │ true │ false Added By │ 31zh6bnhr2tzu7m57huftwdf27re │ 31zh6bnhr2tzu7m57huftwdf27re Added At │ 2026-07-21T20:37:03Z │ 2026-08-05T08:57:02Z Genres │ │ Record Label │ Loma Vista Recordings │ Vingolf Recordings Danceability │ 0.777 │ 0.658 Energy │ 0.823 │ 0.447 Key │ 11 │ 11 Loudness │ -6.507 │ -7.225 Mode │ 1 │ 1 Speechiness │ 0.0784 │ 0.155 Acousticness │ 0.00724 │ 0.836 Instrumentalness │ 0.0207 │ 0 Liveness │ 0.371 │ 0.0752 Valence │ 0.628 │ 0.899 Tempo │ 74.995 │ 87.907 Time Signature │ 4 │ 4
Essentially, every song in Spotify's library has a set of quantitative scores all of which map to its own unique ID and accessible via the Spotify API.
Popularity is Spotify's own score, based mostly on how many plays a song has had and how recent they are. The single and album versions of a song are scored separately.
Genre tags sit on artists, and almost every song inherits them. There are 408 different tags, from "bedroom pop" to "hindi indie", and about 40% of my songs have none at all, Laufey included.
Audio features are Spotify's software estimating how each recording sounds. Most run from 0 to 1:
| Feature | What it measures | Blackberry Marmalade | Lover Girl |
|---|---|---|---|
| Energy | How intense and busy it feels: fast, loud, dense | 0.82 | 0.45 |
| Valence | How positive the music sounds. Sad and angry both score low | 0.63 | 0.90 |
| Danceability | How steady and strong the beat is | 0.78 | 0.66 |
| Acousticness | Confidence that it's acoustic | 0.01 | 0.84 |
| Tempo | Estimated beats per minute | 75 | 88 |
| Speechiness | How much is spoken. The typical rap song lands between 0.33 and 0.66 | 0.08 | 0.16 |
| Liveness | The likelihood it was recorded in front of an audience | 0.37 | 0.08 |
TLDR: They're a computer's guess at how a song feels. Valence hears only the music, so a heartbreak lyric over a bright groove still scores as happy. Tempo detection sometimes halves the real speed: Blackberry Marmalade's 75 BPM may well be 150.
As much as I'd have liked to jump right into the weeds and start tugging, I needed to run some housekeeping first and define what truly counted as "my" music.
I settled on counting a song as a "pick" only if I'd added it to my library myself; either as an addition to the master Liked Songs playlist or to one of the custom playlists I put together. This filtered out auto-generated playlists like Blends or "This is X" as well as any full albums. Including those would have added another 973 songs, 130 of them from the Mario Galaxy soundtrack alone, which would have put a video game well ahead of Kendrick Lamar. I adore Gusty Garden Galaxy, but not enough for it to elevate its entire soundtrack above k.dot.
Another need for housekeeping came from my own inconsistencies in building out my library.
This is 2,643 songs by 2,035 different artists, carrying 408 different genre tags. Some of those genres are sandalwood, vinahouse and polka, each with exactly one song.
Yeah ... I have no idea what vinahouse music is either.
A lot of the mosaic is grey, because Spotify never tagged 40% of these songs. Even so, you can make out a story from the genres that are present, starting with the fact that most of my genre-based loyalties lie with soul, rock & roll, Indian indie and hip hop.
Two things were clear at this stage in the game: This project had now grown beyond Instagram, and there was zero chance I was getting eight hours of sleep.
Split by year, those four loyalties turn into five distinct phases.
My earliest playlists, from December 2017 to May 2018, most of them made on JioSaavn, and the library I brought over from it. Rock (18%) and Indian indie (17%) lead, with hip hop at 14% and electronic at 12%.
I picked very little in 2019 (89 songs, my fewest). What I did pick was soul, jazz rap, lo-fi and bedroom indie. Soul and R&B peaked at 29% in 2020, during lockdown, and jazz sat at 19%.
For two years rock was 2% of my picks. In 2021 it was 20%. Older music came back the same year: 23% of my picks were at least ten years old, up from 5% in 2020, including Stand By Me and songs by The Doors.
Kendrick Lamar was my most-picked artist in both 2022 and 2024. Hip hop peaked at 23% in 2024, the same year I started going to the gym regularly and my Hello from the Jim volumes piled up. Indian indie fell to its lowest point, 7%.
Here, rock took over: 25% in 2025, 28% in 2026, largely driven by my discovery and subsequent obsession with the band Ghost. I also began performing live in 2025, which I'm sure plays at least some small part in the change in taste.
A closer look at my Spotify library as a whole would set both a baseline and potentially show trends that I could later point to Instagram as a contributing factor.
Most descriptions of one's taste in music begin with the artists they listen to. I've always struggled with that. 70% of the artists in my library appear exactly once.
Only Kendrick Lamar clears fifty. Even so, my ten most-picked artists add up to just 7% of the library. The rest is a long tail of single songs I heard somewhere and kept.
There is a world of difference between saving a song to your library and actively listening to it. Fortunately, Spotify had me covered there. The 265 songs in its annual Your Top Songs playlists, plus On Repeat and All-Time Top Songs acted as a strong proxy for what I actually listened to. Comparing them to my picks and with the playlists I follow from friends and curators gives three views of the same person.
My picks rest on four roughly equal pillars: soul and R&B, rock, Indian music and hip hop, each around 15–17%. My plays are lopsided. Hip hop is 15% of what I pick but 24% of what I play. Indian music goes the other way: I save it more than I replay it.
The playlists I follow tell a third story. Jazz is 15% of them and lo-fi 12%, mostly from following Lofi Girl and Tom Misch's huge real good shit. In my own picks, jazz drops to 8% and lo-fi to 2%.
All of this is lovely to know, but doesn't answer the original question. Let's jump in.
(Of course, if you'd like to know more about my learnings courtesy of Spotify's data, click here and read on.)
The overwhelming majority of saves never left the collection. Some songs hadn't been released yet, or hadn't been released to Spotify by the time I write this. Some were indeed songs I already knew or were specific live performances: 42 of the 152 artists on the playlist were in my Liked Songs before I started tracking, including Leon Bridges (liked since 2018), Clairo (2019), Maggie Rogers (2022), Lizzy McAlpine and Phoebe Bridgers (both 2023).
Funnily enough, "save it for later" is literal when it comes to me curating Instagram Bangers: for the 26 tracks first discovered via Instagram, the median wait is ~60 days before they graduate to Spotify. Fastest: Blackberry Marmalade (1 day). Slowest: Supalonely (269), Kiss Me (245), Lover Girl (242).
One artist made the full jump. My first Olivia Dean like came in December 2025. She has more songs on the Instagram playlist than anyone else, eight songs in my Liked Songs, and she's my most-picked artist of 2026. She's also one helluva live performer.
Despite what the funnel looks like, Instagram does still function as a supplementary source of music outside my comfort zone.
Yeah ... about that.
See, it bothered me that I could only confidently attribute less than a fifth of my music to one source. So I ran the same study across Shazam and YouTube, to see if I could glean any more insight.
Shazam was a significantly simpler task. I have my accounts linked, so every song I tag drops into a My Shazam Tracks playlist on Spotify, stamped with the moment I tagged it. Zero need to guess about which came first.
Of the 230 songs I've scanned (shazamed?), 48 made it to my Liked Songs (52 tracks, since a few Shazams match more than one version of a song). The median wait time was 10 days against ~60 for Instagram, and one in five (9 of 48) were liked within a day of tagging.
The music itself is where it parts ways with Instagram. Three in four Shazams were at least two years old when I tagged them, and they skew less mainstream. What Shazam hasn't produced is an Olivia Dean. Jungle and Eminem are the only artists it introduced me to who've since reached four likes, and I'd almost certainly heard of ol' Slim Shady well before Shazam ever entered the picture.
YouTube, on the other hand, is an entirely different beast.
First off, Google auto-deletes one's YouTube history after three months, so the record only goes back to July. That left me with 84 days and 19 liked songs to work with, far too small a sample to reliably compare against Instagram or Shazam. YouTube videos don't carry Spotify IDs either, so I matched them on song title and artist, then went through every overlap by hand.
On paper, 8 of those 19 songs were watched before I added them to my rotation. YouTube was actually where I found two fresh songs: A Tribe Called Quest's Luck of Lucien, courtesy of a British cover band called OMA, and the Middle Aged Dad Jam Band's cover of Just the Two of Us, which I watched six times over a month.
Most of the others came from Instagram. I'd hear a clip, pull up the video, and like the song within minutes: two for Lover Girl, 13 for Good Kid's From the Start, 19 for Karma's Blessing. Karma's Blessing is actually an interesting edge case here, because while I did first discover the song on Instagram, I only started following the band's page because of a completely unrelated show on a streaming service called Dropout.
Where YouTube really shows up is with music I already have. 60 songs I'd liked before July turned up in the history, and 29 of them in at least one live performance: Glastonbury 2007, Letterman, Audiotree, Wacken and most of Ghost's Rite Here Rite Now concert film. Ghost also has the two most-watched songs, Square Hammer (14 watches) and Dance Macabre (12). Another 21 showed up as guitar or bass lessons, tabs and play-alongs.
So YouTube is less a place I find music than one I go back to, either to watch a song played live or to learn it. Its part in discovery is mostly as Instagram's second step.
Each of these studies started tracking on a different date, which makes their numbers hard to line up. So I re-ran Spotify, Instagram and Shazam over the same window, from August 2024 (when I started saving Instagram Bangers) to early October 2026, and credited each like to whichever channel put the song in front of me first. That covers 439 likes. For Spotify, a song counts if one of the 91 playlists I follow added it in the 90 days before I liked it. YouTube only has the one quarter, so it gets a column with that caveat attached.
| Spotify11 likes | Instagram34 likes | Shazam40 likes | YouTubeJul–Oct 2026, 2 finds | All likes439 | |
|---|---|---|---|---|---|
| Put in front of me | 3,063 | 371 | 136 | – | |
| …later liked | 0.6% | 7% | 24% | – | |
| Likes credited | 11 | 34 | 40 | 2 of 19 | 439 |
| New artists | 1 | 20 | 23 | 1 | 221 |
| Median days to like | 23 | 69 | 12.5 | 0 · 30 | |
| In heavy rotation | 45% | 15% | 35% | 1 of 2 | 17% |
| Average popularity (0–100) | 74 | 61 | 55 | 54 · 8 | 60 |
| Median release year | 2024 | 2024 | 2018 | 1990 · 2025 | 2018 |
| Released 2+ years before the like | 27% | 41% | 85% | – | 74% |
| Happy / hype | 73% | 68% | 45% | – | 52% |
| Sad / mellow | 0% | 3% | 2% | – | 14% |
| Energy (0–1) | 0.72 | 0.71 | 0.68 | 0.57 · 0.69 | 0.68 |
| Valence (0–1) | 0.57 | 0.58 | 0.53 | 0.69 · 0.38 | 0.54 |
| Danceability (0–1) | 0.71 | 0.66 | 0.64 | 0.89 · 0.71 | 0.62 |
| Acousticness (0–1) | 0.17 | 0.20 | 0.22 | 0.14 · 0.26 | 0.23 |
| Loudness | −5.9 dB | −6.7 dB | −6.4 dB | −6.3 · −7.3 | −6.9 dB |
| Under 3 minutes | 55% | 38% | 22% | no · no | 26% |
| Explicit | 46% | 38% | 30% | no · no | 27% |
| Top genre (of tagged songs) | Soul/R&B 83% of 6 | Soul/R&B 19% of 16 | Rock 27% | Jazz · untagged | Rock 23% |
Spotify is the firehose. Followed playlists put 3,063 new songs in front of me over those two years and I liked 19 of them, about one in 160. What gets through is the most mainstream music I like (average popularity 74, against 60 overall) and the happiest, and in two years it introduced exactly one new artist. The long view is kinder: count every song a followed playlist held at any point before I liked it, and it's 93 likes. Friends' playlists are something I draw from slowly, over years.
Instagram is where new releases come from. Its finds have a median release year of 2024, they're upbeat, and they take the longest to land, a median 69 days from post to like. They also fade fastest: 15% are in heavy rotation, the lowest of the three. It's the one steady channel, though, with 11 or 12 likes every year.
Shazam is the sharpshooter. One tag in four becomes a like, against one Instagram save in fourteen. It's the quickest (12.5 days), it introduced the most new artists (23), and 14 of its 40 finds are in heavy rotation, twice my average and second only to Spotify's 5 of 11. Its songs are older, with 85% at least two years old when I liked them, less popular, and the loudest of the lot, with rock out in front.
YouTube found two songs in its one quarter. Otherwise it's where I replay and learn songs I already have, and where Instagram clips go to be heard in full.
Over the same two years, Shazam edges Instagram on credited likes, 40 to 34, with Spotify a distant 11. Together they still account for less than a fifth of what I liked. The other 354 have no recorded source, which leaves friends, films, gigs and Spotify's own mixes and autoplay, none of which leave a timestamp.
Some songs I pick once. Others I put into playlist after playlist. Counting how many of my own playlists each song appears in gives a list of anthems. (Blends don't count here, because Spotify fills those automatically with whatever I play most.)
Three of my top five were already in my playlists before I'd ever liked a song on Spotify.
Feel It Still, Way down We Go and LA Nights were all in my library by early 2018, and each has gone into at least nine of my playlists since.
In 2019 and 2020, more than a quarter of what I picked was sad and mellow. In every other year it's 10–18%, apart from the JioSaavn playlists before Spotify (23%). Energy fell to its lowest point too, and acousticness peaked. Since 2021 the energy has climbed almost every year, and 2026 is my happiest year on record: 56% happy or hype, 10% sad.
The typical song in my library came out in 2017, and it's usually about four years old by the time I pick it. That average hides a clean crossover.
In 2019 and 2020, roughly half of everything I picked was brand new and only 5–6% was more than ten years old. By 2025 those lines had crossed: 40% of my picks were back-catalogue, from Aretha Franklin's Respect to Elvis's Jailhouse Rock. The shift lines up with my band starting to play, and with setlists built from songs a crowd already knows.
I've made 107 playlists, and most are for something specific: trips (17), people (20), parties (13), the gym (10), my band (6) and Spotify Blends (8). The other 33 are moods and themes: Sunday Morning, Music to Yearn To, Break Glass in Case Tears are Needed. Only one, The Rock and Roll Remnant, is named after a genre. More than half were built in a single sitting.
Grouping them by what they're for shows a clear split.
Playlists for the gym, parties, the band and Blends are loud, happy and full of hits. Playlists for moods, trips and gifts for one person are quieter, sadder, more acoustic and more obscure.
| Playlist type | Playlists | Energy | Acoustic | Popularity | % new songs |
|---|
In mood and gift playlists, 72–82% of the songs hadn't been in any playlist of mine before. Gym, party and Blend lists mostly reshuffle songs I already know.
That's where things stand in October 2026. Everything above leans on what I saved, and only 265 songs carry real play counts. Spotify's full streaming history and this year's Wrapped should land in November, and the first thing I'll check is whether the songs Instagram brought me are ones I actually play.