The moment half your uploads are made by nobody

If you build products for creators, the ground under your payout rules just moved. AI can now make the content faster than people can, and platforms are scrambling to decide who gets paid and who gets flagged. The numbers are no longer small, and the answers are no longer optional. Let me catch you up.

The flood is already here

This is not a someday problem. Deezer says AI tracks now make up more than half of daily uploads, a monthly average of 90,000 songs a day in June. Rewind eighteen months and that number was 10,000 a day, about 10% of uploads. The curve went straight up: 18%, 28%, 34%, 39%, 44%, then past half.

Video is no cleaner. A study from editing company Kapwing built a fresh YouTube account and tagged 104 of the first 500 recommended videos, about 21%, as AI slop. It also found 278 fully AI channels pulling in 63 billion views and an estimated $117 million a year. When a fifth of what a new user sees is machine-made, your recommendation engine is already spending money on it.

Drawing the line at intent, not tools

Here is the useful part: the platforms getting specific are not banning AI. They are banning cheap output. YouTube spelled out three kinds of "inauthentic" content that lose monetization: generic repetitive videos, off-putting shock bait like fake animal rescues, and AI personas posing as experts on health, money, or law.

Trust and safety chief Matt Halprin put the tension plainly. "AI can actually allow people to make a lot of videos," he said, and some are great. "But that exact same new tool can allow you to make lots of videos really quickly that are very similar." The judgment is not whether AI touched the work. It is whether the finished thing carries a real story, lesson, or point of view.

Deezer draws its own line at behavior. It is pulling AI tracks that went unstreamed for six months or that show fraudulent streams built to skim payouts. The word both are dancing around is payment dilution: every fake stream and every slop video takes a slice of a fixed pot away from the humans your platform is supposed to serve.

The people caught in the dragnet

Your policy will hit the wrong people if you write it lazily. "Faceless" creators, the folks making narrated docs, history explainers, and true crime with real writers and editors behind the screen, look a lot like content farms from the outside. Doctor NOS, an educational YouTuber with 1.8 million subscribers, told THR that creators like him were getting demonetized. Some have started hiring on-camera hosts off Fiverr just to look human, even though YouTube never said a face helps.

That is the trap: enforcement that punishes the honest maker who happens to resemble the spammer. Give real creators a way to show their work. Substack went at it from the reader's side with a Pangram AI detector plus a "How I make this" statement so writers can explain their process. CEO Chris Best named the actual harm: "Platforms that reward fakeness will create a race to the bottom."

Betting the company on humans

The money moves tell you where operators think this goes. Patreon cut 20% of staff, about 93 people, and CEO Jack Conte was blunt that it was not about swapping people for AI. On Decoder he said the opposite pressure was real: if Patreon does not embrace these tools as a product company, "we, as a company, will be dead in three years."

Others are rebuilding the payout math around trust. Taye Diggs launched Microhouse Films on a token model where the platform keeps 25% and the filmmaker takes 75%, plus full access to viewing metrics. He tied it to a Netflix experience where he asked how a film performed and got told the numbers were "under wraps." His pitch to creators is the thing big platforms have been withholding: show them what happened.

The deep cut

The fight is not about detecting AI. Deezer already labels it, Substack ships a scanner, and both admit a detector cannot tell you whether care went into the work. The real lever is your payout rules. YouTube can leave slop up and just make it worthless; Deezer pays out only on tracks fans actually stream. So audit where your money leaks first, then write the enforcement to match. And build an appeal path before you flip the switch, because your policy will flag honest faceless creators on day one, and 21 days of lost income turns your best people into your loudest critics.

Three questions for your team

1. If half our uploads were AI tomorrow, where does that dilute real creator payouts, and can we pay on genuine engagement instead of raw upload volume?
2. What can an honest "faceless" creator show us to prove human work, and how fast can they appeal a wrong demonetization call?
3. Are we hiding performance data the way Diggs got burned by, and would showing creators their real numbers make us the platform they choose?