AI Content Disclosure Rules by Platform: Who Requires What in 2026
- The four rules differ in mechanism: YouTube demonetizes, LinkedIn cuts out-of-network reach, Snapchat removes recommendation eligibility, TikTok tests account-level detection.
- Three of the four describe form — template-like, repetitive, low commentary — rather than whether AI was used. Snapchat is the exception.
- In commerce, regulation outranks preference: China's live commerce measures, effective February 2026, require AI presenters to carry continuous disclosure to consumers.
- The practical rule: disclose where required, and assume the label alone is never the thing being judged.
If you have been trying to hold "the AI rules" in your head as one policy, stop. There are at least four, they use different words, they impose different penalties, and only one of them is really about whether you used AI.
Here is the comparison, with the mechanism each one actually applies.
The four platform positions
| Platform | What it describes | What happens | Origin test? |
|---|---|---|---|
| YouTube | Content that "looks like it's made with a template," repetitive, low educational value, little commentary; plus AI personas advising on health, legal, finance or politics | Not monetizable | Mostly no |
| "Generic and repetitive" content; a user-facing report control for posts that seem like low-quality AI | Reduced visibility outside your own network | No — labels were generic vs original | |
| Snapchat | "Wholly AI-generated" video | Not eligible for Spotlight recommendation | Yes |
| TikTok | Accounts "dedicated to posting AI-generated spam" in politics, financial advice, medical; plus provenance labelling at scale | Detection testing; plus a viewer control for how much AI content people see | Partly — account-level pattern |
Read the fourth column. Three of the four are form tests, not origin tests. LinkedIn's is the most explicit — its human annotators labelled posts generic or original, not human or AI. That design choice tells you what is being optimised against.
Only one of the four penalises AI as such. The rest penalise something AI content is frequently guilty of — and that a human can be equally guilty of.
The penalties are not interchangeable either
This matters for triage, because the same content can be fine on one surface and costly on another:
- Demonetization (YouTube) hits revenue but not reach. Your video still circulates.
- Out-of-network reduction (LinkedIn) hits growth but not your existing audience. Your followers still see you; strangers do not.
- Recommendation ineligibility (Snapchat) hits discovery specifically. Nothing is removed.
- Account-level classification (TikTok) is the most consequential, because it attaches to the account rather than a post — deleting the content does not undo it.
None of the four is a ban. That is the single most under-reported fact about this entire policy wave: the cost of AI-heavy content in 2026 is not removal, it is being seen by fewer strangers.
The layer that overrides platform preference: commerce
Everything above is platform policy, which platforms can change at will. Commerce is different, because regulation sits on top of it.
In China, the Live Commerce Supervision and Management Measures — jointly issued by the market regulator and the cyberspace administration, effective 1 February 2026 — require that where a live commerce operator uses AI-generated person images or video, they must label it in accordance with national rules and continuously indicate to consumers that the image or video is AI-generated. A companion article prohibits using AI to fabricate or spread false or misleading commercial information.
Two things worth extracting even if you do not operate in that market:
- "Continuously" is the operative word. Not a checkbox at publish time — a persistent, visible indication throughout. That is an operational requirement, not a metadata one.
- Commerce is where disclosure stops being optional first. The pattern is consistent: platforms treat general content as advisory and commercial content as mandatory. If you sell, assume the stricter rule.
Chinese platforms have followed the same split. Video accounts on WeChat began requiring AI labelling on commerce short videos from 10 August 2026, with restricted distribution for non-compliance, while non-commerce video remains advisory.
What to actually do
A workable policy that satisfies all four platforms and the regulatory layer:
- Disclose in commerce contexts, always. The cost of labelling is near zero and the cost of not labelling is defined and rising. Do not make this a case-by-case decision.
- Assume disclosure does not protect you. A correct label satisfies a rule; it does nothing about the form tests. Three of the four penalties above apply to labelled and unlabelled content identically.
- Fix the form, not the origin. Every form test names the same properties: template-like, repetitive, low commentary, generic. The remedy is a judgement, first-hand material, and a specific reader — the argument we made in why AI content sounds generic.
- Do not buy detector scores. No platform in this table gates on one, and the research on their reliability is not encouraging. See what AI detectors actually measure.
- Watch the account level, not just the post. TikTok's approach targets accounts rather than videos, which changes what a bad batch costs you. Detail in TikTok's account-level detection.
For anyone running several accounts, the structural implication is the one worth sitting with. Every rule in this table is a similarity rule in disguise. Tooling that helps you publish more of the same thing is pushing against all four simultaneously; tooling that keeps each account's content genuinely independent is at least aligned with them. Neither can supply the thing all four are ultimately looking for, which is a reason for the content to exist.
FAQ
Do I have to disclose AI use on ordinary, non-commercial posts?
It depends on the platform and your jurisdiction, and the answer is changing. The consistent pattern in 2026 is that commercial content carries mandatory requirements while general content is often advisory. Because the advisory side is where rules move first, defaulting to disclosure costs little and removes a category of risk.
Does labelling reduce my reach?
No platform in this comparison states that a disclosure label itself reduces distribution; they state penalties for failing to label where required. The genuinely new variable is audience-side: TikTok is testing a control letting viewers choose how much AI-generated content they see, which shifts part of the cost from enforcement to preference.
What counts as "AI-generated" for labelling purposes?
Definitions differ, and the practical line most platforms draw is whether the processing changes what the viewer believes about who or what they are seeing. Synthetic presenters, voice cloning, face swaps and generated scenes sit clearly inside. Automatic captions, automated cuts and light filters are commonly treated as outside. Confirm against the options in your own publishing interface, since that is where the platform states its own boundary.