Fair Trade AI with Jordan Heather

What Fair Trade AI actually is

AI music tools have to be trained on music, and for years that music was taken without anyone asking. Companies scraped it wherever they could find it, and none of the people who wrote or recorded it ever saw a credit or a cent for it. We built Fair Trade AI because that never sat right with us. It's an opt-in program where you license your own tracks for AI training, on your terms and on the record, and take a share of whatever that licensing earns. You decide to take part, your music goes into a dataset we license to companies building AI audio tools, and you get paid a cut for as long as it's in there. If you ever want out, you're out.

‍

Who can take part, and which tracks qualify

Right now it's open to anyone distributing music through LANDR. Not every track qualifies though, and the main reason is one I'd rather be upfront about than let you find in the fine print: a track is only eligible if you own it outright, both the recording and the publishing. The moment someone else has a claim on it, whether that's a sample you didn't clear, a co-writer you never squared up with or an old label deal, it's off the table. You probably already know which of your tracks those are. We're not being fussy for the sake of it. We can only license music where the person clicking "opt in" has the right to say yes, and that's what keeps the whole thing clean and ethical.

There's one more thing we filter out. We don't license tracks our systems flag as AI-generated. Training new AI on AI-made music defeats the point, since these models need the variety and texture of real human playing, so we leave that material out.

Once you opt in, we curate what goes into each dataset, because every licensing deal calls for something slightly different. Opting in puts your qualifying tracks in the pool; it doesn't mean all of them are included in every dataset. To be very clear about what you're agreeing to, you're giving LANDR permission to include those tracks in licensed datasets that train AI audio models, in return for a share of the money. You're not selling or signing away ownership of your music. It's closer to licensing your name onto a run of merch with a clothing company. They get to use it on agreed terms for a while and it stays yours the whole time.

‍

The money: how the 25% works

A quarter of the net licensing revenue goes to the artists in the program. It was 20% when we started, and we pushed it up to 25% because giving more back to you was the whole point. Your share is worked out after the costs of the deal come out, including any third-party charges involved in running and licensing the dataset.

The share itself is split by how much you put in. Every qualifying track you have in a dataset is a slice of it, so if your track is one of ten thousand, you earn 10/10,000th of that dataset's 25%. It's plain proportional maths, and you can see your own breakdown in your LANDR account.

A dataset only earns what the deals behind it earn, and the market for licensed training data is still young, so spread across a lot of tracks any single payout is small for now, and I'm not going to pretend otherwise. What it has going for it is that it promises recurring payments alongside your distribution earnings, and it climbs as we sign more deals and more money comes through. As of today we've paid out over $153,000 to artists, and that figure keeps moving up week by week. It's the real number, and we'd rather show it than tuck it away.

‍

You own your music.

This is opt-in from the first click to the last, and you can leave any time you like. The part worth being straight about is what leaving does. Opting out stops your music going into any future dataset from that point on, but it can't undo the past. Picture a covers band that's already learned one of your songs: you can tell them not to learn any more of yours and they'll stop, but you can't reach inside their heads and make them forget the one they already play. A trained model works the same way. If a track of yours has already gone into a dataset that's already trained a model, that's baked in, and there's no pulling it back out. So opting out protects everything from here on, but it doesn't rewind what's already happened, and you deserve to know that going in rather than after.

None of this touches your ownership. Your music stays yours. Licensing it for training doesn't sign it away or put anyone between you and your masters.

‍

The uncomfortable questions

So what happens to my music? It goes into a licensed dataset, and AI companies use that dataset to train audio models. In practice, "training" means the model studies your track for its musical qualities, the melody, harmony, rhythm, timbre and tempo, and learns the patterns underneath them.

Do the AI companies get my actual songs? No, and this is the bit people most often get wrong. Your recording isn't sitting inside the model and can't be pulled out or played back from it. What the model keeps is closer to what a musician takes away after studying a thousand records: a feel for how they're built, not a copy of any one of them. Nobody's downloading your track from the far end of this.

What do the partners get, then? Music they're cleared to use and that's been tagged properly for training, with the people who made it paid for it. That's the thing that's been missing from AI music until now, and it's what they're paying for.

‍

Why this beats the alternative

Staying out of AI was never really on the table. Your music has been training these tools for years, most of it scraped, and nobody asked you or sent you a penny for it. So the real choice isn't "AI or no AI." It's whether your music gets used with your say-so and a share of the money, or used anyway with neither.

We understand this type of program can feel scary, and as a releasing musician myself, I would feel exactly the same. I take comfort in the fact we’re doing the right thing by fellow musicians to get fairly compensated for their artistry. 

So take your time with it. Opt in if it feels right, leave it if it doesn't. Either way, now you know what you'd be saying yes to.