A channel with 40,000 subscribers, publishing faceless AI narration videos about pensions and savings accounts, is exactly the kind of channel that now has a monetization problem. Not a takedown problem. A payment problem. The videos can stay. The revenue may not.
YouTube’s VP of Trust and Safety, Matt Halprin, used a Creator Insider interview to explain how the updated Partner Program rules treat AI, and the shape of it is narrower and more awkward than the “YouTube bans AI slop” headlines suggested.
One word changed, and it matters
The policy language moved from “repetitious content” to “inauthentic content”.
That is a small edit with a large consequence. Repetition is countable. A system can measure how similar your last forty uploads are to each other. Authenticity is a judgement about whether a person contributed something worth paying for, and there is no counter for that.
Halprin was clear that YouTube is agnostic about which tools creators use. Generative video is not banned. Three things are restricted inside the Partner Program, according to YouTube’s updated monetization policy:
- Generic, repetitive video, especially obvious content farming
- Off-putting or distressing material, with animals in peril given as the example
- “AI personas” dispensing advice on serious subjects such as personal finance
The third one catches more legitimate businesses than the first two. A synthetic presenter explaining debt consolidation, insurance excess or tax deadlines is now a monetization risk whether or not the advice is accurate.
These rules sit apart from the community guidelines every uploader follows. Breaching them does not usually remove the video. It removes the money.
What this actually asks you to prove
Because the test is authenticity rather than repetition, the useful question is not “how much AI did we use” but “what would we point to if someone asked what a human contributed”.
For a small business channel that is a short list, and most of it is free:
- An original script written from something only you know: your own pricing, your own callouts, your own customer questions
- Original footage or original data, even ten seconds of it, even a photograph of the actual job
- A named human somewhere in the chain, in the description or on camera, rather than an invented presenter
- A reason this video exists that is not “this topic gets views”
Take a plumbing company running a how-to channel with AI voiceover over stock footage. Nothing about that is prohibited. But if the script is a rewritten version of the top three search results and the presenter is a synthetic person called Dave, the channel is sitting in two of the three restricted categories at once. Replacing the stock footage with phone video from real callouts, and the invented Dave with a real engineer’s voice, costs an afternoon and moves the channel out of the risk zone entirely.
Where this goes wrong
The obvious risk is over-correcting. Plenty of channels will strip AI out of workflows where it was doing no harm, lose their publishing rhythm, and end up worse off than if they had changed nothing. Editing, captioning, thumbnail variants and translation are not what this policy is about.
The less obvious risk is the enforcement gap. YouTube has moved to a standard that automated systems cannot measure directly, which means decisions will be inconsistent and appeals will matter. If a channel is a meaningful revenue line, keeping a record of what was human-made is not paranoia, it is the only evidence available when a review goes the wrong way.
This may not be worth much attention at all if AI is a small part of how you publish and a person is visibly involved. The channels that should be reading the Creator Insider interview line by line are the ones where nobody could name the human who made the last video.
The part that is easy to miss
YouTube has not drawn a line between AI content and human content. It has drawn a line between content someone made and content someone generated, and those are not the same distinction. A fully AI-produced video with an original argument, real data and a genuine reason to exist sits on the safe side. A human-edited video assembled from other people’s search results does not necessarily.
If you are auditing a channel this week, sort by “could I defend this one in a sentence” rather than by which tool made it. The wider set of platform labelling and disclosure rules is covered in our guide to AI content policy for marketers.
Editor’s note: This area changes quickly, so check the latest platform policy before making compliance decisions.