This blog used to have three separate posts about AI in music. They were written in 2025, they were breathless, and they quoted a future that mostly hasn't shown up. This one honest version replaces all three.
The short answer: AI has changed the tools a lot, changed the competition somewhat, and changed the actual job of building a career very little. The details matter, so let's go through them.
Where it genuinely helps
The most useful AI tools for an independent artist are the boring ones. They take a slow, technical task and make it fast.
Stem separation. Pulling vocals, drums, or bass out of a mixed track used to be close to impossible. Now tools like Moises or LALAL.AI do it well enough for practice tracks, live backing tracks, and remix sketches. (Only use it on music you have the rights to.)
Cleanup. Noise reduction, de-essing, fixing a vocal recorded in a bedroom with a fridge running. Hours of fiddly work, now often minutes.
Mix and master assistants. Tools like iZotope Ozone or LANDR analyze your track and suggest EQ, compression, and loudness settings against a genre target or a reference. Good for a demo, a quick reference, or catching a problem you can't hear anymore because you've listened 400 times.
Mock-ups. Want to know if strings would work on the bridge before you pay string players? You can sketch it in minutes, decide if the direction is worth it, then bring in real players with a clear brief.
Words around the music. Drafting a pitch email, a bio, show announcement copy, or a caption you'll then rewrite in your own voice. The admin side of a music career is mostly writing, and a first draft is the hardest part.
What these have in common: you stay the one making the decisions. The tool saves time. It doesn't supply taste.
Co-writing: a spark, not a songwriter
The honest version of AI co-writing looks like this. You're stuck on a bridge for two weeks. You ask a tool for ten options. Nine are wrong for the song. One has a chord move you wouldn't have tried, and it knocks something loose, and you end up writing a bridge that is neither yours from before nor the tool's. That's useful. It's the same thing a good co-writer does in a room when they say "what if it went here?"
It's also where the limits show up fast. A tool can produce a competent lyric about heartbreak in seconds. It can't produce yours, because it doesn't know the specific night, the specific street, the detail only you would think to include. Those details are why people connect with songs. Generic lyrics are now free and unlimited, which makes the specific ones worth more, not less.
The skill that matters more now is curation. Knowing which of the ten options is right, and why, and when the answer is "none of them." That's taste, and taste is still the job.
Fully generated songs: what the flood means for you
Tools that turn a text prompt into a finished, sung, produced song are now good enough that Deezer, one of the major streaming services, has said fully generated tracks passed half of its daily uploads in mid-2026, while still making up only about 1 to 3 percent of actual listening. Most of it gets uploaded. Very little of it gets heard.
That sounds scary for an independent artist. My honest read: it does make the pool of "competent background music" effectively infinite. If your plan was to win on being competent, that plan got harder. But a listener who follows an artist, saves their songs, comes to a show, and buys a shirt isn't choosing based on competence. They're choosing a person. Generated tracks can fill a lo-fi study playlist. They don't build a fanbase, because there's nobody there to be a fan of.
So the flood mostly raises the value of the things a machine can't fake: a recognizable voice, a point of view, a face, a live show, a relationship with the people who listen.
How labels and A&R scout now
The 2025 version of this post described "AI A&R" systems that spot the next star months before any human. Reality is plainer. Labels, publishers, and managers do lean heavily on data dashboards to decide who to look at. Those dashboards track things like streaming growth, how many listeners come back, short-form video traction, and where an artist's audience is concentrated. A human still listens and decides.
The useful part for you: they're mostly looking at signals you can see yourself. Not total streams, but whether people stick. Saves. Streams per listener. Whether listeners return after a release week. Whether your audience is clustered in a few cities where you could actually play. An artist with an engaged 800-listener base can look more interesting than one with 40,000 passive listeners.
And the systems are better at spotting fakes than they used to be. Bought streams and bot plays tend to show up as obvious spikes with no saves, no follows, and no return listening. Spotify charges labels and distributors a per-track fee when it detects flagrant artificial streaming, and distributors can pass that cost, or a takedown, on to you. Fake numbers don't get you scouted. They get you flagged.
If you want to see what your own numbers say, the way a manager would read them, the Read is free and takes a minute. Paste your Spotify artist link.
Where it's mostly hype
Be skeptical when you hear any of these:
"Our system predicts hits." Nobody can reliably predict which song will connect. Data can tell you which of your songs is getting saved more. It can't tell you the future.
"Get discovered by the algorithm." There's no algorithm you can be discovered by in the abstract. Recommendation systems amplify songs that real listeners already respond to. The listeners come first.
"Smart targeting guarantees streams." Promo services love adding the word AI to their pitch. A guarantee of streams is a warning sign no matter what technology is attached to it. If you're unsure about an offer, paste it into the free promo offer checker before you pay.
"One click mastering means you're done." Assistants get you close. For a release you care about, a good human ear still catches what the preset can't.
Rights and disclosure: the basics
This isn't legal advice, and the rules are still moving, so check current terms before you release anything. But the basics are fairly stable.
Human authorship is what gets protected. The US Copyright Office's position is that purely machine-generated material can't be copyrighted, and that prompts alone generally don't make you the author. What you write, perform, select, arrange, and meaningfully change can be protected. Keep your drafts, session files, and voice memos. They're your record of what you made.
Read the tool's terms. Some generators only grant commercial rights on paid plans, and some keep rights to what you make. Know what you're agreeing to before a generated element ends up in a release.
Disclose honestly to your distributor. Many distributors now ask whether a release contains AI-generated material, and some restrict fully generated tracks. Answer truthfully. Getting it wrong can mean a takedown. Spotify supports AI disclosures in song credits through the industry's DDEX standard and has started showing them in a song's credits when the distributor passes them through, so listeners can increasingly see how a track was made.
Never clone someone else's voice. Releasing music that imitates another artist's voice without permission breaks platform rules and, in some places, the law. It will get removed, and it can get your account removed.
Split sheets are for people. If you write with humans, do a split sheet the same day. A tool isn't a co-writer and doesn't get a share.
Your own work and training. If it matters to you whether your music trains someone's model, read the upload and distribution terms of the services you use.
Being open with your fans about how you work is a good default too. Most listeners don't mind tools. They mind feeling tricked.
What doesn't change
Strip away the tools and the career still runs on the same few things it always did.
Songs people want to hear twice. Every tool above makes it easier to finish a song. None of them make the song matter. That's still you.
Fans, one at a time. The artists who last build real relationships: replying to comments, running an email list, knowing the 50 people who show up to every show by name. No tool fakes that, and listeners can tell.
Showing up. Releasing consistently. Playing live. Pitching your release to Spotify editorial through Spotify for Artists at least a week before release day. Doing the unglamorous weekly work of checking your numbers and picking the one thing that matters next.
Honest judgment. More tools mean more noise, more promo pitches, more people selling shortcuts. Knowing which advice fits your actual situation matters more, not less.
That last one is why Musuni exists. It's built to be your manager: it reads your real numbers, tells you plainly what they mean, and helps you do the next useful thing, without promising streams you can't get honestly. If you're not sure where you fit yet, start with the free artist-type quiz. It takes two minutes, and it's a decent lens on how you make music and reach people.
The technology will keep changing. The job of making something people care about, and then finding those people, won't.