Do You Have to Label AI Videos? TikTok, YouTube and Instagram Rules

The first time I uploaded an AI-generated clip to TikTok I did the responsible thing and typed "AI-generated content" at the end of the caption. I felt fine about it. I was also, technically, not compliant — because a line of caption text is not the disclosure any of these platforms actually ask for, and none of them read it.

Every major platform now has a real machine-readable AI label, and each one is set in a different place: a field in the upload API, a toggle in the composer, or metadata baked into the file itself. This is what each one wants as of August 2026, what happens when you skip it, and the one that still has no clean answer.

Why the caption line does nothing

Platform AI labels are structured data. They live in a field, and that field drives a badge rendered on the video, feeds the platform's own provenance systems, and in some cases changes how the clip is distributed. Free text in a caption is not read by any of that machinery — it is read by humans, some of the time, after they have already watched.

That matters practically, not just legally. On the platforms that render a badge, an unlabelled AI clip that later gets classified as AI by automated detection is treated as a labelling failure rather than a labelled post. You want to be the one who declared it.

Keep the caption line anyway. It costs nothing, it is honest, and it covers the case where someone screenshots the video out of the app. Just never let it be the only thing you did.

TikTok: a field in the API, not a checkbox you can forget

TikTok's AI-generated content flag is is_aigc, and if you publish through the Content Posting API it has to be sent explicitly. The detail that catches people out is that it belongs in both calls — the initialisation step and the publish step. Send it in one and not the other and you can end up with a clip that uploaded fine and carries no label.

Nothing sets it for you. I publish through an automation layer and had to add it by hand in both places; the SDK default is off. If you post manually in the app instead, the equivalent is the "AI-generated content" switch under Content disclosure, and TikTok will sometimes prompt you for it when its own detection fires.

TikTok also reads C2PA Content Credentials on upload and will apply the label automatically if the file carries them — which is the most reliable route of all, because it survives re-uploads and does not depend on you remembering.

YouTube: one boolean, buried in the status object

On YouTube the field is status.containsSyntheticMedia, set at upload through the Data API. In YouTube Studio it is the "Altered or synthetic content" question in the upload flow's details step.

YouTube's rule is narrower than most people assume. It applies to realistic synthetic content — material a viewer could mistake for a real person, place or event. Obvious animation, heavy stylisation and clearly unreal imagery are outside it. Our own clips are photoreal, so we set it on every upload; if yours are stylised illustration you may genuinely not need it, and over-declaring has no penalty either way.

The consequence of getting it wrong is graduated rather than dramatic: YouTube may apply the label itself, and repeat failures can affect monetisation standing. That is a slow problem, which makes it easy to ignore until it is not.

Instagram and Facebook: the awkward one

This is the gap. Meta's Graph API has no field for the AI-info label on organic posts. You cannot set it programmatically at publish time the way you can on TikTok and YouTube, which means an automated pipeline posting straight through the API has no clean route to a compliant label.

There are two real answers. The first is to schedule rather than publish directly, and set the AI toggle in Meta's own composer when the post is queued — which is what we do, and it happens to match our scheduling policy anyway. The second, and the better long-term one, is to embed C2PA or IPTC provenance metadata in the file before upload. Meta reads both and applies "AI info" automatically.

The metadata route is worth the setup cost because it is the only one that is not a human remembering. It also travels with the file: the same master that goes to Meta carries its provenance to TikTok, which reads the same standard.

What this costs you in reach

The fear that keeps people from labelling is that the badge suppresses distribution. Across roughly two dozen labelled posts a month on each platform, I have not been able to see it. Our labelled clips sit inside the same wide performance spread as everything else, and the variance between a good hook and a bad one dwarfs anything attributable to the badge.

That is not a controlled experiment and I would not present it as one — the honest version is that if there is a penalty, it is smaller than the noise in short-form performance. What I am confident about is the asymmetry: the downside of an unlabelled clip that gets auto-classified is real and compounding, and the downside of labelling is, at worst, small.

Where it genuinely helps is trust. In the AI video niche the audience assumes AI anyway, and declaring it reads as confidence rather than apology.

The working checklist

Per clip, before it goes out: set is_aigc in both TikTok calls; set status.containsSyntheticMedia on the YouTube upload; schedule the Meta posts and flip the toggle in the composer, or ship provenance metadata in the file; and keep the caption line as a backstop rather than a solution.

Then do the thing that actually makes it stick: put it in the pipeline, not in your head. Ours sets the two API fields automatically now, and the only manual step left is the Meta toggle — which is exactly the one that gets missed, which is why the metadata route is next on our list.

If you are building the pipeline itself, the TikTok guide covers the posting side and the business guide covers the wider setup.

Looking for something else? Browse all 72 AI video guides in one list.

I run both of my brands on this stack and I publish the real numbers — spend, cost per purchase, which clips flopped — inside the community every week. If you want the working version rather than the sales page, join us on Skool.