Start with the part that surprises people: no major platform bans AI-generated video. TikTok, Instagram and YouTube all allow it, all distribute it, and all show plenty of it. What every one of them requires is that realistic synthetic content be labelled, and the meaningful change of the last two years is not in the rules themselves but in the enforcement. Disclosure used to be an honor system. In 2026 it is a detection system, and the honor part is optional.
What is AI UGC covers the short version of this. This post is the long one: what each platform actually does, dated, because any article about platform rules starts expiring the day it is published.
From asking to detecting
The mechanism underneath everything below is provenance metadata. The C2PA standard lets a tool sign the file it produces with Content Credentials, a machine-readable record of how the file was made. When a platform reads those markers, it does not need to ask you whether the video is synthetic. It knows, at upload, regardless of which toggles you did or did not switch.
The practical consequence is worth sitting with: whether your content gets labelled is increasingly decided by your tooling, not by your honesty. The remaining choice you own is whether the label appears with your cooperation or despite it.
TikTok reads the file
TikTok has integrated C2PA Content Credentials since January 2025. If it finds provenance markers in an upload, it applies its AI-generated label itself, and it has labelled well over a billion videos that way. The manual toggle still exists and is still worth using, both because not every tool writes credentials and because switching it yourself reads better than being caught by the scanner.
On the paid side the rule is firmer: advertising that contains synthetic media has to be disclosed, and undisclosed AI content in ads is a rejection reason rather than a suggestion. The organic feed forgives more than the ads manager does.
Meta labels and classifies
Meta runs the same play across Facebook and Instagram with its AI info label, which became mandatory for AI-generated content in May 2026. It reads provenance metadata like TikTok does, and it also runs its own classifiers that infer origin from the content itself, which means stripping metadata does not put you back in control; it only removes your say in the matter. On the advertising side disclosure is a requirement, and undisclosed AI content has become a meaningful share of ad rejections.
YouTube asks at upload
YouTube took the questionnaire route. Since March 2024, the upload flow in Creator Studio asks whether the content contains realistic altered or synthetic material, and answering yes attaches an altered or synthetic content label. For most videos it sits in the expanded description; for sensitive topics such as news, elections, health and finance it is displayed prominently on the player. Shorts are covered like everything else, and repeatedly failing to disclose is treated as a policy violation rather than an oversight.
The EU put a floor under all of it
Since August 2026, the transparency obligations of the EU’s AI Act apply: AI-generated content has to be marked in a machine-readable way, and realistic synthetic depictions of people, deepfakes in the regulation’s language, have to be visibly disclosed. Platforms rarely build region-specific compliance for something this structural, which is why the EU rule matters even for accounts with no European audience. It hardens the direction every platform was already moving in.
What to actually do
The playbook that falls out of the above is short and mostly free:
- Prefer tools that write Content Credentials. Provenance in the file means the labelling happens correctly without you thinking about it again.
- Switch the platform toggles on anyway. Belt and braces, and it costs one tap.
- Never present a synthetic person as a real customer. A generated presenter can voice a question or demonstrate a product; a fabricated testimonial is the one use every platform, and now a regulator, treats as deception.
- Re-check the policies quarterly and date what you find. This is the fastest-moving corner of platform rules, and that includes this post.
Disclosure stopped being a courtesy the day detection started doing the enforcing.
The labels themselves cost almost nothing in a feed where synthetic content is ordinary. What costs is the strike for hiding what the platform could see anyway. If you want the per-platform detail on formats and specs rather than rules, the guides for TikTok, Instagram Reels and YouTube Shorts carry that side.
