The AI label and C2PA: when it is required and how to apply it
This is the most important topic in the whole AI stage, because getting it wrong can end in the video being removed.
[OFFICIAL] The rule itself is short: AI-generated content depicting realistic scenes must be clearly disclosed. TikTok may take down such content when it is not labelled.
Start with the definition. [OFFICIAL] TikTok defines AIGC as video, image, audio, text or LIVE content fully or partially generated, edited or synthesised with artificial intelligence. That includes content that simulates, alters or recreates people, scenes or events — and content that is entirely fictional or presented in a particular artistic style (painting, cartoon, animation).
| Content | Label required? | Why |
|---|---|---|
| AI-generated realistic person or scene | Yes — mandatory | Disclosure of realistic AIGC is required |
| AI voice or a clone of your own that sounds real | Yes | Audio falls within the AIGC definition |
| Synthesis using a real person's face or voice | Yes — and you need their permission | Impersonation is prohibited; realistic AIGC must be labelled |
| Obviously cartoon or animated style | The requirement eases when it is not realistic — label it if unsure | The requirement centres on realistic scenes |
| TikTok's own AI effects | No — labelled automatically | The platform applies the label itself; no manual step |
| AI-assisted editing (captions, pause removal) | No — no scene is generated | An editing tool does not simulate a scene |
| AI-written script, your own footage | No | The picture and audio are real |
How to apply the label — [OFFICIAL] there are two routes:
- The toggle (the main route): open the post settings page from the "Post +" button and turn on the "AI-generated content" switch. A label reading "Disclosed by creator as AI-generated" appears in the bottom-left corner
- Manual disclosure: on-screen text, a sticker, a watermark, or a clear note in the description
[OFFICIAL] Content made with TikTok's own AI effects is labelled automatically — no extra step.
Practical advice: use the toggle. People may not read a note in the caption; the toggle sits on the video itself and satisfies the rule unambiguously.
C2PA Content Credentials — the label arriving without you doing anything.
[OFFICIAL] TikTok works with C2PA (the Coalition for Content Provenance and Authenticity) and is the first video-sharing platform to implement Content Credentials. Content Credentials attach provenance metadata to content, and TikTok can read it and apply the AIGC label automatically.
The practical consequences:
- Some AI tools attach that metadata to their output. Upload such a file and the label may appear on its own
- This is not a penalty — it is a transparency mechanism
- Trying to strip the metadata is pointless: TikTok [OFFICIAL] uses several detection methods, including invisible watermarking
[PRACTICE] The right approach is to accept it up front: if you use AI, apply the label yourself. There is no verified data showing the label hurts performance — while the damage from an unlabelled video being removed is certain.
The hardest red line: impersonating real people.
[OFFICIAL] TikTok's Integrity and Authenticity guidelines prohibit content that harmfully misleads viewers or impersonates others. That covers AI content misleading about real-world events, fake authoritative sources, fabricated crisis events, and false depictions of public figures.
This is separate from the labelling question: a label does not make such content permissible. Even labelled, synthesising a real person without their permission is prohibited.
The practical rule: do not use AI to make a real, recognisable person speak. Not with a label, not without one.
🛠 Practice task
Run the labelling decision tree on the AI content you last posted (or plan to) and write down the outcome.
Then open the post settings page and find the "AI-generated content" toggle — seeing where it lives once is what stops you forgetting it later. Use a test video to see how the label appears in the bottom-left corner.
It is done when the decision tree's verdict is written down, you know where the toggle is, and you have checked whether the AI tools you use attach C2PA metadata.
📚 Sources and documentation
- AI-generated content guidelinesofficialtiktok.com
The AIGC definition, the labelling requirement, and how to switch the toggle on.
- Supporting responsible, transparent AI-generated contentofficialtiktok.com
C2PA, automatic labelling and the detection methods.
- Integrity and Authenticity guidelinesofficialtiktok.com
The prohibition on misleading AI content and impersonation — a layer separate from labelling.
- C2PA — Content Credentialsofficialc2pa.org
The provenance metadata standard itself.