As AI video becomes indistinguishable from real footage, labelling is becoming both a legal requirement in some places and an expectation from audiences and platforms. Clear labelling protects viewers from being misled and protects you from accusations of deception. This guide explains when to label, how to do it and the technical tools that help.
Why labelling matters
- Trust: audiences increasingly want to know what is real. Honest creators keep their trust.
- Platform rules: major platforms require disclosure of realistic synthetic content.
- Legal requirements: several jurisdictions have introduced or proposed transparency rules for AI-generated content, especially deepfakes and political material.
- Misinformation prevention: clips travel without context. A label helps prevent your creative work from being misused as fake news.
When labelling is needed
Rules vary, but a strong general principle is: label when a reasonable viewer could mistake the content for real footage of real people, places or events. This includes:
- realistic people who do not exist presented in a documentary or news style;
- real people shown saying or doing something they did not, which also requires consent;
- realistic depictions of events that did not happen;
- altered footage of real events;
- synthetic voices of real people.
Clearly fantastical, animated or stylised content, such as dragons, cartoon characters or obviously surreal worlds, is less likely to require a label under platform rules, although many creators still mention AI use voluntarily.
How to label
Platform tools
YouTube, TikTok, Instagram and Facebook provide settings or labels for AI-generated or altered content. Use them when your content meets their criteria. Some platforms also apply labels automatically when they detect provenance metadata in uploaded files.
In the description or caption
A short, clear note works: “Made with AI video tools”, “AI-generated visuals”, “This scene is an AI reconstruction, not real footage.”
On screen
For content likely to be clipped and reshared, an on-screen note or small watermark is the most robust label, because captions are often lost when videos are reposted. For realistic reconstructions, a brief title card at the start can be appropriate.
In credits
For films and longer works, list AI tools in the credits alongside other tools and contributors.
Provenance metadata and content credentials
Industry standards such as content credentials, based on the C2PA specification, attach information about how content was created and edited to the file. Some AI tools add this metadata automatically, and some cameras, editors and platforms read or display it. Benefits include:
- a verifiable record of AI involvement;
- automatic labelling on platforms that support it;
- protection for your genuine work, by proving origin.
Metadata can be removed by re-encoding or screenshots, so it complements visible labels rather than replacing them. Some AI providers also use invisible watermarks embedded in the pixels, which are designed to survive some editing.
Wording examples
| Situation | Example label |
|---|---|
| Fantasy short film | “Visuals created with AI video tools. Story, edit and sound by [name].” |
| Historical reconstruction | “AI reconstruction for illustration. Not archival footage.” |
| Product ad with AI scenes | “Scenes generated with AI. Product shown as sold.” |
| Avatar presenter | “Presented by an AI avatar.” |
| Consented likeness | “Digital likeness used with [name]’s permission.” |
What labelling does not do
A label does not make harmful content acceptable. It does not replace consent for using a real person’s likeness, and it does not make illegal content legal. Labelling is about honesty, not permission. See Deepfakes, Consent and Likeness.
Does labelling hurt performance?
Many creators worry that labels reduce views. Platforms generally say labelled content is not penalised for the label itself. Audiences are increasingly comfortable with AI-made creative work when it is honest; what they dislike is feeling tricked. A good video with a label performs better over time than a deceptive one that gets reported or debunked.
A simple labelling policy
- Use platform AI labels whenever content is realistic and could be mistaken for real.
- Mention AI tools in descriptions for all AI-assisted videos.
- Add on-screen notes for realistic reconstructions and anything likely to be reshared out of context.
- Keep provenance metadata where your tools provide it.
- Never remove labels or watermarks from other people’s AI content to pass it off as real.
Labelling for clients
If you produce AI video for clients, agree on labelling in the brief. Brands carry the reputational risk if an ad is later revealed to be AI without disclosure, and many now have internal policies. Recommend clear, simple labels and document the agreement. It protects both of you.
Labelling in different formats
Different formats need slightly different approaches:
- Short-form vertical clips: use the platform’s AI toggle and a short caption note. A small on-screen tag in a corner works well because these clips are often reshared.
- Long YouTube videos: use the upload disclosure, mention AI in the description and, for documentaries, add a note at the start explaining that visuals are reconstructions.
- Client ads: follow the ad platform’s policy and the brand’s rules; a short line in the ad copy is common.
- Festival films: follow the festival’s disclosure rules and list AI tools in the credits.
- Website and background loops: a line in the page footer or credits is usually enough.
A real-world scenario
A creator makes a realistic “1920s street scene” reconstruction for a history channel. Without a label, a clip of it goes viral on another platform as “rare footage from 1923”, and history enthusiasts start citing it as real. With an on-screen note reading “AI reconstruction, not archival footage”, that misunderstanding is far less likely, and the creator’s reputation stays intact. A few words of disclosure prevent a problem that would be very hard to correct afterwards.
Frequently asked questions
Do I need to label stylised animation?
Often not under platform rules, but voluntary disclosure is good practice.
Will my video be labelled automatically?
Possibly, if it contains provenance metadata a platform detects. Do not rely on it; label yourself.
Is a hashtag enough?
A hashtag helps, but platform labels and clear text are more reliable and visible.
