In January 2026 YouTube deleted or hid the videos of at least eighteen channels publishing AI slop. Not demonetised, removed. One of them had more than 1.2 billion views.
Every large platform changed what it pays for during the year, and they all moved the same way: authorship gets paid, volume does not. If your format involves producing a lot of similar things quickly, this is the year that arrangement ended.
What follows is what each platform actually said, with dates, so you can check the originals rather than take our word for it, then a self-audit, because the policies name formats rather than individual videos, and the useful question is whether yours is one of them.
YouTube: the Inauthentic Content policy#
In July 2026 YouTube renamed its Repetitious Content policy to Inauthentic Content and set out three categories that cannot be monetised. The clarification rolled out on 16 July and applies to everyone in the Partner Programme.
Generic or repetitive content#
Material that looks template-made or feels repetitive to a viewer: low educational value, little commentary, and AI-generated content built from generic or unoriginal templates that gives the impression of mass production.
The operative word in the policy is template. It is not about how the content was made or whether AI was involved. It is about whether the videos differ only in the subject slotted into an identical structure.
Unsatisfying or off-putting content#
A new category and the vaguest of the three. Content leaning heavily on emotionally manipulative formulas, mimicking existing formats, or built to shock people into watching.
This is the one to watch, because a lot of perfectly sincere content uses emotional hooks. The distinction being drawn appears to be between a hook that leads somewhere and a hook that is the entire product.
AI personas giving advice#
AI-generated representations of people discussing sensitive subjects, finance, legal matters, healthcare, medicine. YouTube has said explicitly it does not want to incentivise this, and it is the clearest of the three: a synthetic presenter giving financial or medical advice is out, regardless of whether the advice is any good.
The distinction that got lost in the coverage#
AI used as a tool remains fully monetisable. Script drafts, editing, translation, disclosed voiceover, thumbnails, research, all fine. What is excluded is AI replacing authorship, where the finished video contains no real human input and could have been produced identically a thousand times over with a different topic.
A great deal of panicked commentary in July read the policy as “AI content is demonetised”. It is not, and building your workflow around that misreading would cost you the productivity gains for no benefit.
Enforcement runs on three strikes: a warning, a 90-day suspension, then removal from the Partner Programme.
Sources: TechCrunch, Tubefilter, Engadget, YouTube’s own policy page.
And separately, higher thresholds#
Announced August 2026, effective 1 February 2027. New creators need 8,000 qualified watch hours over 365 days, or 20 million qualified Shorts views over 90 days. Keeping Shorts revenue requires an ongoing 10 million qualified views per 90 days. Existing members are not affected by the entry thresholds.
In exchange, new earning routes that do not depend on advertising: Shopping bonuses, brand deal incentives, trend bonuses, and Premium Lite globally at a 60% net revenue share against 30% on standard Premium. Platform ad revenue works through what those numbers mean in practice.
Disclosure of realistic AI content#
Separate from monetisation. You must declare when AI has been used to generate or meaningfully alter photorealistic content, making a real person appear to say or do something they did not, altering real footage, or generating realistic scenes that never happened. There is an “AI use” toggle in Studio at upload. See disclosing AI use.
Is your format at risk? A self-audit#
The policies describe patterns, not videos, so the honest test is structural. Six questions:
- Could someone swap the topic and produce your next video without you? If the structure, script shape, voice, visuals and pacing are fixed and only the subject changes, that is the template pattern the policy names.
- Does any individual video contain a judgement only you would make? An opinion, a decision to cut something, a personal experience, a conclusion the source material does not state. If not, there is no author in the output.
- Is the hook the whole product? If the first five seconds promise something the remaining two minutes do not deliver, that is the shape of the new “unsatisfying or off-putting” category.
- Is a synthetic presenter giving advice on money, law or health? This one is explicit and there is no ambiguity to work with.
- Is your upload rate only possible because nobody is really making the videos? Forty videos a month from one person is a signal, not an achievement.
- Would a viewer feel they had learned something from a specific person? This is the underlying question all the others approximate.
Format archetypes that are exposed under these rules: faceless narration channels reading scraped text over stock footage, story-aggregation channels reworking forum posts, AI-voiced list videos, quiz and compilation channels with no commentary, and synthetic-presenter finance or health channels. Some of these had been quietly profitable for years. They are not a business any more.
Meta#
Meta published updated original-content guidance in March 2026, under the heading of rewarding original creators, clarifying how work gets recommended in Feed and Reels and cracking down on unoriginal and repackaged material. It is also testing improvements to content protection tooling, including impersonation detection with an easier reporting route.
At the same time it launched Creator Fast Track, paying $1,000 a month to creators with at least 100,000 followers on Instagram, TikTok or YouTube and $3,000 a month to those with over a million, to post on Reels and Facebook, with the usual waiting period before monetisation waived.
Read those two together and the strategy is legible: pay real creators to come, stop paying people who repost. In April, product tags moved into Reels and Meta declared “the era of link in bio is over”. In May, paid subscriptions launched across Instagram, Facebook and WhatsApp, alongside professional plans at $14.99 and $49.99 a month offering verification, impersonation protection and more prominent placement.
The practical exposure on Meta is different from YouTube’s. It is less about AI and more about reposting: content lifted from elsewhere, watermarked clips from other platforms, and compilations of other people’s work. If your Reels strategy is reuploading TikToks, that is the behaviour the originality rules target.
Sources: Meta’s announcement, Tubefilter on Creator Fast Track, Tubefilter on product tags.
Spotify#
More than 75 million tracks removed as spam over the past year. The filter targets behaviour rather than AI use as such: flooding the platform with high volumes of low-effort tracks, duplicate content under multiple artist names, manipulating titles and metadata for search visibility, and inflating stream counts.
AI-assisted music is permitted. Voice clones, mass uploads, stream fraud and misleading metadata are the enforcement risks. Spotify is also building an AI disclosure standard with DDEX, so artists can state which elements involved AI: vocals, instrumentation, post-production. It introduced a Verified badge for human artists in April 2026.
Sources: ABC News, Music Business Worldwide, Tubefilter.
TikTok#
The US divestiture completed on 22 January 2026: 45% of US operations went to an investor group including Oracle, Silver Lake and MGX, with ByteDance retaining around 20%. Oracle is the designated security partner, responsible for data audits and for replicating and retraining the recommendation algorithm under US jurisdiction.
Creator fund, advertising, TikTok Shop and payouts are substantially unchanged. What is less predictable is reach, while a retrained algorithm learns from US behaviour. If you have deals paid on performance rather than a flat fee, that is worth revisiting, see contract clauses that cost you money.
Sources: eMarketer, Romano Law.
What safe AI use looks like#
Since the line is real but not obvious, here is roughly where it falls based on what the platforms have published.
Clearly fine: AI for research and first drafts you then rewrite; editing, captioning, transcription and translation; thumbnail generation; disclosed voiceover of your own script; B-roll and illustration that nobody would mistake for real footage; upscaling and cleanup.
Fine with disclosure: photorealistic generated scenes, altered real footage, synthetic voice of yourself, anything that could mislead a viewer about something that happened.
Not monetisable regardless of disclosure: templated output at scale with no authorial input; a synthetic presenter giving financial, legal, medical or health advice; content whose only variation between episodes is the topic.
Never: a real person’s likeness or voice used without permission to say things they did not say.
The reliable test is not “did I use AI” but “is there a person’s judgement in the output, and could a viewer tell”. If yes to both, the tools you used to get there are your business.
If you have been demonetised#
There is an appeals process on every platform and it is worth using, but be realistic about what it does. Appeals succeed when a specific rule was misapplied to a specific piece of content. They rarely succeed when a channel’s whole format falls within a policy, because in that case the system is working as intended.
Read the actual policy text, not a summary. Identify which named category you have been placed in. This matters because the argument for each is different.
Appeal on that category specifically. “This video contains original commentary at 2:14, 5:30 and 8:45 and the structure differs from my previous uploads in the following ways” is an argument. “I work hard on my channel” is not, however true.
Do not re-upload the same thing. It compounds rather than resets.
If the format matches the policy, the appeal will not fix it. Unwelcome, and better to know in week one than in month four. The work is changing the format, and the honest version of that is usually making fewer things with more of you in them.
We cannot help with individual appeals and have no route into any platform. Nobody outside the platform does, whatever they charge.
The part worth taking away#
Read all four platforms together and the same rules emerge.
Volume formats are finished as a monetisation strategy. Not throttled, explicitly excluded, by name, on the largest platforms.
AI as a tool is safe; AI as the author is not. This distinction is stable enough to plan around and every platform drew it in the same place.
Reposting is the Meta-specific risk, as templating is the YouTube-specific one and mass uploading is the Spotify one. Different words, same underlying position.
Entry got harder and originality pays better. Thresholds rose while new payments appeared for shopping, brand deals and trends. Platforms are redistributing towards people they consider authors, and the redistribution is happening whether or not you agree with their definition.
All of it points at the same conclusion this site keeps arriving at from different directions. A channel that can be demonetised by a policy update is a business with a single point of failure, and 2026 demonstrated repeatedly that the failure is not hypothetical. The answer is not to police your format more carefully forever. It is to have income that does not depend on a platform’s opinion — a product, a service and an audience you can reach directly.
Policies in this area changed several times during 2026 and will change again. Check the platform’s own current policy text before making a decision, and please tell us if something here has moved: [email protected].