On this page
All three platforms draw the same basic line: if AI makes something look or sound real that isn't — a person saying words they never said, an event that didn't happen — you must label it. Using AI to write a script, fix the colours or generate a cartoon doesn't need a label. The platforms differ on sound: Meta wants a label on realistic AI voices; YouTube and TikTok only when the voice imitates a real person.
The rules side by side
What AI did in your video | YouTube | TikTok | Instagram, Facebook, Threads | EU AI Act (professional use) |
|---|---|---|---|---|
A real person appears to say or do something they didn't (face swap, voice clone) | Required | Required | Required | Required |
Footage of a real event or place altered | Required | Required | Required | Required |
A realistic scene that never happened | Required | Required | Required | Required |
Human-sounding AI narrator, not a real person's voice | Not required | Not required (generic text-to-speech is exempt) | Required | Not required |
AI-generated music as the focus | Required | Not named in the guidelines | Required | Not required |
Clearly unrealistic visuals (animation, anime, fantasy) | Not required | Not required | Not required | Not required |
Beauty filter, colour correction, background blur, auto-captions | Not required | Not required | Not required | Not required |
AI wrote the script, title or thumbnail idea | Not required | Not required | Not required | Not required |
When in doubt, the test all three platforms use is the same: could a viewer mistake this for something that really happened, or for a real person?
YouTube: realistic content and AI music
YouTube requires disclosure when content "makes a real person appear to say or do something they didn't do", "generates a realistic scene that didn't actually occur", or "creates music that's the main focus of the video". It doesn't require it for clearly unrealistic content or for edits that are "primarily aesthetic" — beauty filters, caption creation, and even cloning your own voice for a voice-over or dub.
How to label: in YouTube Studio, under Attributes, set "AI use" to Yes when uploading. The YouTube Data API also has a field for it, containsSyntheticMedia. The label appears in the expanded description, or on the player itself for photorealistic content and AI music. YouTube may also add a label on its own when content was made with its own AI tools, carries C2PA content credentials, or is detected by its systems.
TikTok: realistic people and scenes, and real voices
TikTok's Community Guidelines (2026 update) require a label when "AI or significant edits" show realistic-looking people or scenes — a face replaced with someone else's, someone made to say something they didn't, a person or object added or removed in a misleading way, or AI audio that mimics a real person's voice. They name three cases that need no label: small edits like colour correction, reframing or cropping; artistic styles such as anime; and generic text-to-speech narration that isn't the recognisable voice of a known person.
How to label: use TikTok's AI-generated content label when posting, or a clear caption, sticker or watermark. TikTok also labels content made with its own AI effects automatically, and reads C2PA content credentials from some other tools.
Watch out: a label doesn't make every AI video allowed. TikTok doesn't allow AI content that misleads on matters of public importance — fake news footage, crisis events, or false endorsements by public figures — label or not.
Instagram, Facebook and Threads: realistic video and realistic audio
Meta asks people to disclose "a photorealistic video or realistic-sounding audio that was digitally created or altered", and "may apply penalties" if they don't. Realistic-sounding audio is the difference from YouTube and TikTok: a natural-sounding AI narrator counts on Meta's platforms even if it isn't anyone's real voice. Meta also labels images on its own when it detects industry-standard indicators that they are AI-generated.
How to label: use the "Add AI label" option when sharing a post or reel.
The EU AI Act: a legal duty, not a platform rule
From August 2, 2026, Article 50(4) of the EU AI Act requires anyone who uses an AI system in a professional capacity to disclose deepfakes they publish: image, audio or video content that resembles existing people, objects, places or events and would falsely appear authentic. It applies to brands, agencies and creators who earn from their content; purely personal, non-professional use isn't covered. For evidently artistic, creative, satirical or fictional work the disclosure can be lighter, as long as it doesn't hamper the work. A generic AI narrator or AI music doesn't make a video a deepfake on its own.
This is a summary to help you decide, not legal advice. If your business publishes realistic AI content at scale in the EU, check the details with a lawyer.
Check your own video
Tick what AI did in your video in our free AI disclosure checker and it shows, for YouTube, TikTok, Meta and the EU AI Act, whether a label is required — with the rule that triggered it. It uses the same rules as the table above.
When a video is marked as AI-generated in Uplonic, the AI label is sent with the post to TikTok, Instagram and Facebook Reels. YouTube's API field isn't connected yet, so for YouTube set "AI use" in Studio after the video is published. The other per-network rules for your videos are in our social media video specs overview.
Sources & methodology
How we got this: from each platform's own help pages and community guidelines and from the text of the EU AI Act, re-checked on October 17, 2026. Platform rules change; the date on each source is when we last read it. This article is not legal advice.
Disclosing use of altered or synthetic content — YouTube Help, accessed October 2026
Videos resource (status.containsSyntheticMedia) — YouTube Data API, accessed October 2026
Community Guidelines — Integrity and authenticity — TikTok, 2026 update, accessed October 2026
Labeling AI-Generated Images on Facebook, Instagram and Threads — Meta Newsroom, February 2024, updated April 2025
