What tagging your music as AI actually costs

You ticked the AI box and watched the release do nothing. Or you did not tick it, and now every comment that says "this sounds like Suno" reads like a countdown.

The tag is not really about you. It was built for a volume problem, and the volume is worth seeing before deciding what to do about your own release.

The tag exists because of the numbers, not because of taste

Deezer publishes its own figures. In an announcement dated 21 July 2026, it reported "90,000 AI-generated tracks per day", which at peak in June 2026 was "over 50% of all new music uploads".

Two more numbers from the same page matter more than the first one. Fully AI-generated music accounts for "between 1-3%" of actual streams, and "up to 85% of the streams generated by fully AI-generated tracks were in fact fraudulent in 2025".

So the picture is not an audience quietly switching to machine music. It is a flood of uploads that almost nobody plays, most of whose plays are fake. Deezer's response is to keep those tracks out of algorithmic recommendations and editorial playlists, and to take down tracks used for fraudulent streams along with any that have gone six months without a stream.

Both of your options are bad, and that is the actual complaint

A working musician's post in July 2026 argued that honest AI creators lose to the ones who hide it. The reply at the top states the trap better than the post did:

"If they are honest and mark their work as AI produced, Youtube algorithm and other distribution channels reduce their reach, hurting their chance of success. If they don't mark their work as AI, they may get wider reach, but are always at [risk] of getting demonetized, reported, or for people to complain that they're using AI and not disclosing it."

Another reply names the design flaw: the platforms punish the people who do the right thing, then act surprised when the less scrupulous do not follow them.

Neither reply is a policy document. Both describe what people are experiencing, which is the only evidence anyone has about reach.

The checkbox has one setting and your song has many

Here is a verse I wrote, with no model involved:

[Verse 1]
My father's name is on the mailbox still
I paint around it every second spring

And here is what came back when the same verse went through a model asked to improve it:

[Verse 1]
My father's memory lingers in the air
I feel his shadow with me everywhere

One of those is mine. One of them is nobody's. If I write the first pair, hand them to Suno to sing, and release the result, the disclosure field says the same thing it says about a track where a prompt did everything.

A widely read July 2026 post proposing a human-created-music-only toggle ran into the same objection within a few replies: what about AI assisted rather than AI generated, given that plenty of musicians use it as a tool rather than a replacement. Nobody in that thread had an answer, and no platform has published one either.

Note

That post is a proposal, not news. Its author says so in the first line. Treat the toggle as a thing people want, not a feature anyone has shipped.

Disclose it, and be specific where you can

The metadata field is binary, but everywhere you write in your own words is not. A description that says which parts are yours, in one sentence, costs nothing and answers the question the tag cannot.

The distributors' own rules point the same way: disclosure at upload, proof of commercial rights on request, and penalties that land on the account rather than the track. That last part is the argument that settles it. A reach penalty applies to one release. A removed catalog takes every honest release with it.

What disclosure will not do for you

It will not restore the reach. Nobody has published a number for what the tag costs, and the people reporting it are describing their own channels.

It will not protect a track whose streams look bought, since that is what the detection was built for in the first place.

And it will not tell anyone which half of the work was yours, which is the part worth writing down yourself.

Before your next upload, write one sentence for the description saying what you did and what the model did, and keep it with the lyrics.

We are taking this decision on our own release. Echoes of Tomorrow comes out on Friday with its AI use disclosed, and the artist bio says it in one sentence: the words were written by a person, line by line, and the voice and the arrangement are AI-generated. Whether that costs us the reach described above, we will find out in public.

This is what the platforms and the threads said in September 2026, and platform policy moves faster than anything else here. It is not legal advice.