After an AI video is finished, clients often ask: do we have to label it as AI when we upload? The answer is: it depends on the content. This page sets out the principles the platforms share. Rules change, so check the platform's own help page before publishing.
The principle all three platforms share
YouTube, Meta (Facebook, Instagram) and TikTok all require a label when AI-generated or AI-altered content "looks like a real person, event or scene". The test is "could someone mistake it for real", not "was AI used".
- Usually needs a label: realistic-looking people speaking or acting, events that appear to have really happened, imitations of a real person's voice or face, content that looks like real news footage.
- Usually doesn't: clearly animated or stylised visuals; AI used only for colour, background removal, subtitles or scripting; obviously unrealistic effects.
How each platform labels
- YouTube: an "altered or synthetic content" option in the upload settings; ticking it shows a label on the video.
- Meta: an AI-content option when posting; the platform may also add an "AI info" label automatically based on detection.
- TikTok: an "AI-generated content" label option when posting; files carrying content credentials may be labelled automatically.
Option names and locations change; follow the interface at the time of posting.
Ads need extra care
Ad rules are stricter than post rules. Meta requires disclosure of digitally created or altered content for certain ad categories (social issues, elections, politics); all platforms restrict AI imitations of real people and potentially misleading imagery. Have whoever runs the ad account check the current ad policy before launching.
Our advice
- If in doubt, label. Unlabelled content that gets reported may be removed or restricted.
- Don't use AI to create what looks like a real customer testimonial or real news footage.
- Content that needs a real person (the owner, staff, customers) is filmed; AI handles scenes and scenarios.
- At delivery we tell you which segments are AI-generated, so you can decide when posting.
Frequently asked questions
Does labelling reduce reach?
The platforms currently say the label itself doesn't affect recommendations; reach depends on how the content performs. This may change; follow the platform's statements.
Only some shots are AI. Do we label?
If those shots meet the "looks real" test, we recommend labelling.
Does an AI voice-over need a label?
If the voice imitates a specific real person, it must be labelled and the person must consent. Ordinary synthetic narration isn't mandatory on most platforms, but labelling is safer. See how to do a voice-over for a video.
