Platforms now force a decision: should you tag a post as AI? The risk is concrete—mistakenly realistic AI results can be removed, damage audience trust, or trigger disputes. Use a single, easy test to decide in seconds, then apply the platform controls for TikTok, Meta (Instagram/Facebook), and YouTube when required.
The one test that should decide your label
Ask: did AI create or meaningfully change something a viewer could mistake for real footage or audio?
If yes, label the video. That includes photorealistic clips generated from scratch, lifelike synthetic people or avatars, invented events presented as real, and edits that make a real person appear to say or do something they did not.
If AI only improved presentation without changing what actually happened—auto-captions, color correction, noise reduction, upscaling, routine cuts, or script/thumbnail polish—you generally don’t need an AI label. Those edits increase production value without altering the underlying facts of the clip.
Borderline edits: use caution
These common workflows sit between clear cases and require judgment:
- AI dubbing or voice cloning: If a generated voice could be mistaken for the original speaker, disclose it.
- Generated B-roll or synthetic backgrounds: Replacing missing shots with AI-generated footage can mislead about what occurred—label when it implies events that didn’t happen.
- Generative object removal or reconstruction: Removing or reconstructing elements that change the story should be disclosed.
- AI versions of people: Using a synthetic or AI-generated version of a person on camera needs disclosure when it could be read as genuine.
When a platform’s guidance is vague, return to the test above: if the result could pass for reality, label it.
How major platforms handle disclosure
Each app exposes disclosure controls in different places and surfaces labels differently. Use the platform controls at upload; platforms can also add labels automatically when they detect provenance metadata.
TikTok. TikTok requires labels for AI-generated content containing realistic images, audio, or video. Enable it during upload (Next → More options → AI-generated content). TikTok may add labels automatically when creators use its AI effects or when a file carries provenance metadata. The platform can remove unlabeled realistic AI content and says enabling the label won’t affect distribution if the video follows Community Guidelines.
Meta (Instagram and Facebook). Meta exposes an “AI info” tag in a post’s advanced settings and applies labels automatically when it detects provenance data. Meta treats fully AI-generated posts and lightly AI-edited ones differently: fully generated images or videos receive visible labels, while light edits have the AI info tucked into the post menu rather than shown prominently. Disclosure is still required when the threshold is met, but visibility varies.
YouTube (including Shorts). At upload YouTube asks whether the video contains “altered or synthetic” content. YouTube’s examples of edits that do not require disclosure include beauty filters, color work, captions, upscaling, sharpening, cloning your own voice for a voiceover, and thumbnail or script help. Disclosure is required for photorealistic scenes that did not occur, edits that make a real person appear to say or do something they did not, or footage altered to misrepresent an event. Highly realistic AI content and music receive visible labels; less overt AI edits are noted in the expanded description. YouTube states that accurate disclosure does not penalize reach or monetization.
Quick internal workflow for teams
Turn the test into a three-step checklist editors can follow before publishing:
- Does the clip or audio depict something that didn’t happen or a real person saying/doing something they didn’t? If yes → label.
- Does the edit create a lifelike synthetic person or photorealistic scene? If yes → label.
- If neither, but the edit involves dubbing, generative fill, or synthetic b-roll, review the asset and default to labeling when in doubt.
Log the tool and asset provenance when you produce content. Preserve or export provenance metadata (Content Credentials) so platforms can surface correct labels automatically and your team retains an audit trail.
Train editors on where to add the label at upload (TikTok: More options; Meta: Advanced settings; YouTube: the synthetic/altered content prompt). If you use third-party publishing tools, confirm they preserve platform disclosure fields.
Risks and what to monitor
Unlabeled realistic AI content can be removed by platforms. Platforms may also apply labels automatically if files carry provenance metadata. Because Meta sometimes hides AI info for light edits, consider adding an explicit caption note when visibility matters for authenticity or brand trust.
Platforms state that disclosure itself shouldn’t reduce reach or monetization when content otherwise complies with rules. Still, labels can change viewer perception and engagement; monitor analytics to see if and how audience response shifts.
Policy and detection are evolving. Small changes in how provenance is detected or how labels display can alter best practices—so track platform updates and adjust your checklist accordingly.
Practical next steps
Adopt the single test above as your default, add a short upload checklist for each platform you use, and retain provenance metadata where available. When in doubt, label: transparency protects your audience, reduces the risk of takedown, and preserves brand credibility.
