How to Label AI-Generated Videos in 2026: EU and Platforms

To label AI-generated videos in 2026 you do two things. You disclose in the caption and on screen, and you set the platform's own AI label when you post natively. Since August 2, 2026 the EU AI Act asks anyone who publishes a deep fake for work to say so. This is not legal advice, just the public rules as of November 2026.

What does EU AI Act Article 50 require from August 2, 2026?

Article 50 applies from August 2, 2026. It puts one duty on the people who publish AI video and another on the tools that make it. The text is in Article 50 of Regulation (EU) 2024/1689, with the date set by Article 113.

The Act calls a publisher a deployer: anyone using an AI system under their authority. Article 50(4) covers deployers whose AI system makes or alters content that is a deep fake. They "shall disclose that the content has been artificially generated or manipulated". Article 3(60) defines a deep fake: AI-made or altered content that looks like real people, places, things or events and could pass for real.

A few details change how you apply that.

• Timing and form. Article 50(5) wants the information to reach the viewer at the latest at first exposure, in a clear and distinguishable way. A note in a profile bio fails that test; a caption line that shows before "more" or an on-screen line passes it, and so does a platform label on the player rather than in an expanded description.
• Creative work. When a deep fake is part of an evidently artistic, creative, satirical or fictional work, the duty shrinks to disclosing that generated content exists, without hampering the work.
• Text. AI-generated text published to inform the public on matters of public interest has to be disclosed too, unless a human reviewed it and takes editorial responsibility. A LinkedIn post about your product usually sits outside that duty. It does not inform the public on a matter of public interest. Where one does, the exception needs a person who checked the substance and takes editorial responsibility, not a spell check.

For the tools, Article 50(2) puts a duty on providers. Systems that generate synthetic audio, image, video or text must mark their outputs "in a machine-readable format and detectable as artificially generated or manipulated". Systems already on the market before August 2, 2026 have until December 2, 2026 to comply. Platforms read that mark when they label a video on their own.

The Commission added a voluntary Code of Practice on marking and labelling AI-generated content on June 10, 2026, with EU icons. Fines for ignoring Article 50 go up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher, under Article 99(4).

What does the Act not require?

It does not ask you to label every video that touched an AI tool. Article 2(10) excludes natural persons using AI in a purely personal, non-professional activity, and a brand account is professional.

A synthetic presenter who looks like nobody in particular is still inside it when a viewer could take the clip for filmed footage. The Commission's Article 50 FAQ reads "existing" to include people who could plausibly exist. The working rule for a brand: if the clip could pass for filmed, disclose. If it is obviously animated, the definition does not reach the pictures, though a cloned real voice over them is still a deep fake.

The Act says nothing about label size or wording; the platforms do.

How does each platform label AI-generated video?

Every platform has a switch you set, a label it sets, or both.

Platform	Label you set	Label the platform sets	Primary source
TikTok	"AI-generated content" toggle at upload, or a sticker or caption	Auto-label on content made with TikTok's own AI effects or carrying C2PA Content Credentials	TikTok Help, Sep 19, 2023, May 9, 2024
YouTube	"AI use" question under Attributes in Studio at upload	Label in the description, or on the player for photorealistic content and AI music; auto-applied on C2PA or YouTube-made content	YouTube Help
Instagram, Facebook, Threads	Disclosure tool when posting realistic AI video or audio	"AI info" label on detected content; AI-edited content shows it in the post menu	Feb 6, 2024, Apr 2024
X (Twitter)	Armed-conflict disclosure for revenue-share creators; a "Made with AI" composer toggle, reported March 2, 2026	"Made with AI" indicators for EU users (July 2026 Media Literacy Action Plan); reach restriction or removal of deceptively altered media under the Authenticity policy	X policy, Engadget, Mar 3, 2026
LinkedIn	None in the help center; disclose in the post text	C2PA icon on media signed with Content Credentials, showing whether AI was used	LinkedIn Help
Pinterest	None; disclose in the Pin description	"AI modified" label in closeup, from metadata and classifiers, with an appeal	Apr 30, 2025

TikTok

TikTok requires labels on realistic AI images, audio and video. It added a creator-side toggle on September 19, 2023. Since May 9, 2024 it auto-labels uploads carrying C2PA Content Credentials. It labels content made with its own AI effects too. Set the toggle yourself on anything realistic rather than counting on an automatic label. TikTok also runs its own detection, but nothing guarantees it catches your clip. TikTok says the setting does not affect distribution for a post within its guidelines.

YouTube

YouTube wants a disclosure in four cases. AI makes a real person appear to say or do something they did not. AI alters footage of a real event or place, or generates a realistic scene that did not happen. Or AI creates music that is the video's main focus. Non-realistic content and minor edits are exempt: filters, captions, upscaling, script help, cloning your own voice for a voiceover.

The label sits in the expanded description, and can appear on the player too for photorealistic content and AI music. The help page says the disclosure alone does not limit reach or monetization. After upload, the AI disclosure survey under Attributes in Studio is where you correct a wrong automatic label. Creators who keep skipping it risk a manual label, removal or suspension from the Partner Program. The October 1, 2026 originality change for Shorts is a separate matter, covered in YouTube Shorts ideas.

Meta: Instagram, Facebook, Threads

Meta set out its rule on February 6, 2024. Anyone who posts lifelike video or audio that AI made or changed must use its disclosure tool. Meta may apply penalties when they do not. In September 2024, content that AI only modified moved its label into the post menu.

X (Twitter)

X's Authenticity policy lets it restrict the reach of, or remove, deceptively altered media. The first disclosure requirement arrived on March 3, 2026. Creators in the revenue-sharing program must disclose AI-generated videos of armed conflicts, and a first offense costs 90 days out of the program.

A "Made with AI" toggle in the composer was reported rolling out on March 2, 2026, spotted by an app researcher. X's help center has since referred to "Made with AI" indicators for EU users in its Media Literacy Action Plan. It does not document the toggle. Set it where you see it, and write the line into the post text as well.

What does a working disclosure routine look like for a brand?

One habit covers the Act and every platform rule above. Decide at script time whether the clip could pass for filmed. Then disclose on every copy of it.

1. Classify each video before posting as realistic (a presenter, voice, place or event a viewer could take as real) or evidently animated.
2. For realistic clips, write one disclosure line into the caption template, something like "Presenter and voice generated with AI", above the fold and on every channel. The line survives cross-posting where metadata often does not.
3. Put a short on-screen line in the first seconds of the clip. It is seen at first glance and survives a re-upload.
4. Set the platform switch when you post natively: TikTok's AI-generated toggle, YouTube's AI use question, Meta's disclosure tool. A scheduled post never passes through that upload screen, so the caption line and the on-screen line are its disclosure. Where the platform wants its own switch, as YouTube does for its four cases and Meta for lifelike video, post that clip natively.
5. For dubbed videos with a cloned voice, treat the audio as realistic and say so, even when the pictures are real.
6. Keep a record of which tool made which clip, and when.

The label costs you less reach than you fear. YouTube and TikTok both say the disclosure alone does not cut distribution; the other platforms say nothing on reach either way. What they punish is a skipped disclosure, with the penalties listed above. The ad side is in what are UGC ads. The episodic side is in microdramas.

The caption is the disclosure that reaches every channel from SocialFaktory. Write the line once and it goes out with every copy of the clip through scheduling. Set up an account and put the line in your caption template before you post anything.

In one sentence: EU brands must disclose realistic AI video at first exposure, so say so in every caption and on screen and set the platform's switch when posting natively.