The AI label on YouTube, Instagram and TikTok: when you need it and how to add it (November 2026)

If you post videos made with AI, every platform has its own rule on when to say so. And since August 2, 2026, there's the EU AI Act as well. Here's what YouTube, Instagram, Facebook and TikTok ask, how to add the label and what you risk without it, from the official pages we read on November 3, 2026. This isn't legal advice: it's what the texts say.

The short version

• The question to ask: does the video look real and show people, places, events or voices in a way that didn't happen? Then you need the label, everywhere.
• Usually not needed for animation, obvious fantasy, filters and small touch-ups, or a script written with AI.
• Watch the voice: on Instagram and Facebook, a realistic AI voiceover is enough to require the label.
• Disclosing doesn't cost you reach: YouTube and TikTok both say the label doesn't change how videos are distributed.

YouTube

When it's required. For realistic content generated or significantly altered with AI: a real person saying or doing something they never did, a real event or place altered, a realistic scene that never happened. It's not required for animation, obvious fantasy, filters, captions, upscaling, or a script, title or thumbnail written with AI. Not for your own cloned voice either.

How to add it. In YouTube Studio, while uploading, find the "AI use" question among the video's attributes and answer "Yes". Apps that post for you use the API field containsSyntheticMedia.

Where it shows. Since May 27, 2026, on Shorts the label sits over the video, and on long videos right below the player; for animated or lightly edited content it stays in the description.

Automatic labels. YouTube adds the label itself on videos made with its own tools (such as Veo and Dream Screen), on videos with C2PA metadata, and, since May 2026, when it detects significant photorealistic AI that wasn't disclosed. In some of these cases you can't remove it.

What you risk. For creators who consistently don't disclose, removal of videos or suspension from the Partner Program. Disclosing, on the other hand, doesn't change recommendations or earnings.

A different rule worth knowing. Since July 15, 2025, YouTube doesn't monetize "inauthentic" content, meaning mass-produced or repetitive videos. The AI label has nothing to do with it, but anyone making videos with AI should know.

Instagram and Facebook

When it's required. Meta asks for the label on photorealistic video and realistic audio that was digitally created or altered, before you post it. That includes a Reel with a realistic AI voiceover and a song with an AI voice. A cartoon-style landscape doesn't need it. For images it isn't mandatory, but Meta labels them if it detects AI.

How to add it. On Instagram, while creating the post, with the "Add AI label" toggle. An AI-made profile is flagged from "Edit profile". Apps that post for you use the is_ai_generated field, on Instagram since June 22, 2026 and on Facebook Page Reels.

Automatic labels. Meta reads the C2PA and IPTC signals that many AI tools put into files. The label's name has changed more than once ("AI info", then "AI content"), so don't be surprised if you see different wording.

What you risk. Meta mentions possible penalties without saying which. AI-made profiles that aren't disclosed can be left out of recommendations.

TikTok

When it's required. The guidelines in force since September 24, 2026 ask for the label on AI-made or significantly edited content showing realistic people or scenes: a swapped face, words put in someone's mouth, things added or removed to mislead, an AI voice imitating a real person. It's not required for small touch-ups, artistic styles such as anime, or a generic synthetic voice.

How to add it. When you upload, before posting, under "More options" there's the "AI-generated content" toggle. A clear caption, sticker or watermark also counts. Apps that post for you use the is_aigc field; but if an app sends the video as a draft to your TikTok inbox, you add the label yourself when you finish the post.

Automatic labels. TikTok labels videos made with its own AI effects and those with C2PA content credentials, and since November 2025 it also adds an invisible watermark. The automatic label can't be removed.

What you risk. Unlabeled content can be removed, shown less or labeled by TikTok's team. Mind the opposite too: labeling a video as AI when it isn't goes against the terms.

What the EU asks since August 2, 2026

Article 50 of the AI Act asks anyone who uses AI professionally to disclose that a deepfake is generated or manipulated, clearly, at the latest when someone first sees it.

What a deepfake is, for the Commission: content that closely resembles existing or plausible people, places or events and could pass as real. In the examples of the July 20, 2026 guidelines, a famous influencer generated with AI in an ad is a deepfake, and so is a realistic avatar of a CEO. Talking mice in an ad, the AI voice of a made-up character, or an AI background behind a real product are not.

Worth knowing:

• for clearly artistic, satirical or fictional works, a light disclosure that doesn't spoil the work is enough;
• the platform's own label can be enough if it's clear; a note hidden in a menu is not;
• the obligation doesn't cover personal use and isn't retroactive: content made before August 2, 2026 doesn't need relabeling;
• there's an optional EU "AI" icon, in the code of practice published in June;
• fines go up to €15 million or 3% of worldwide turnover (for small businesses, whichever is lower).

In Italy

Law 132 of 2025, in force since October 10, 2025, made it a crime to spread AI-falsified images, videos or voices without consent when it causes unjust harm: one to five years. The authority overseeing the AI Act in Italy is ACN, the national cybersecurity agency. And influencers that AGCOM considers "relevant" (at least 500,000 followers, or a million average monthly views) must flag filters or software that could mislead, with wording such as "edited video" or "filters".

What to do

• Ask the question for every video. If it looks real and shows people, places, events or voices in a way that didn't happen, add the label on every platform. On Instagram and Facebook the voiceover counts too.
• If an app posts for you, check it sets the right field (containsSyntheticMedia, is_ai_generated, is_aigc). In HypeFox, for example, the AI label is on by default for every video, on YouTube, Instagram and TikTok.
• Disclosing costs little. It doesn't change distribution, and if you don't, the label may arrive on its own, sometimes for good.
• If you post professionally in the EU, flag deepfakes visibly from the start of the video.
• No faces or voices of real people altered without their consent.

Sources

Official pages read on November 3, 2026 (the Instagram and TikTok help pages that now require a login were read from archived copies from October and June 2026): YouTube, disclosing AI content, the labels from May 27, 2026, C2PA, the API and Partner Program policies; Meta, AI labels on Instagram, Meta help, the 2024 announcement and the API changelog; TikTok, community guidelines, help on AI content, the Content Posting API and invisible watermarks; EU, AI Act, Article 50, Q&A, guidelines and code of practice; Italy, Law 132/2025 and AGCOM's influencer code.