Do you have to label your AI video? Disclosure rules for US creators

One thing worth knowing before any of the detail. The tool you generate with almost certainly doesn't label anything — the obligation lands on whoever publishes, which is you.

What actually needs a label?

The test is realism plus mistakability. Not whether AI touched the file.

Synthetic media is the term all three platforms use. It means realistic content that was generated or meaningfully altered by AI.

The right-hand column is worth internalising. Over-labelling isn't a safe default, it's noise, and it trains your audience to ignore the label that matters.

Does your generation tool label anything?

Usually not, and this is the assumption worth checking first.

Hexcoded states it plainly: it adds no AI label and no watermark to output, and any AI-disclosure labelling required where you publish is your obligation as the publisher. Its acceptable use policy puts the same point more bluntly — the platform provides the tool, not clearance of your campaigns.

Other tools differ, and some embed provenance metadata even when they add nothing visible. The practical consequence is that your generation tool and your disclosure obligation are separate systems. Finishing a video doesn't discharge anything.

What does each platform require?

YouTube

Disclosure is required when AI has meaningfully altered or generated photorealistic content. You set it at upload.

Non-compliance risks a label being applied for you, content removal, or Partner Program suspension.

TikTok

TikTok requires people to label realistic AI-generated content, and layers several mechanisms to enforce it: creator labelling tools, its own detection models, and C2PA Content Credentials — a cross-industry metadata standard that lets platforms recognise AI-generated files.

It's also rolling out invisible watermarking, which only TikTok can read and which is harder to strip when content is re-uploaded or edited elsewhere. That began with content made using TikTok's own AI tools and uploads carrying C2PA credentials.

TikTok said in July 2026 it had labelled more than 3 billion videos as AI-generated, up from 1.3 billion in November 2025.

Meta

Facebook, Instagram and Threads use an "AI info" label, applied both from creator disclosure and from Meta's own detection. Detection runs on industry-standard provenance signals, including the AI-generated fields in the C2PA and IPTC standards.

The FTC layer

This runs in parallel and isn't satisfied by any platform label. Commercial-relationship disclosure and AI disclosure answer different questions — one tells the audience the post is paid, the other tells them the content is synthetic. If a post is both sponsored and synthetic, treat it as needing both.

Where this falls short. This is the least precisely sourced part of the post. We haven't found an FTC statement addressing the interaction between the two obligations directly, so treating them as independent is the conservative reading rather than a stated position. Platform disclosure mechanics also changed several times during 2026 — check before a significant delivery.

Does the EU AI Act apply if you're a US creator?

Reaching EU viewers is the trigger. Not being established in the EU.

Article 50 became applicable on 2 August 2026. Obligations apply from that date to all in-scope systems regardless of when they entered the market, so content published before then doesn't need a retroactive label.

Deployers using AI to create realistic deepfakes must disclose that the content was artificially generated or manipulated — unless it's had substantive human editorial review with a named person taking editorial responsibility.

A voluntary Code of Practice on marking and labelling AI-generated content offers a compliance route, including a published icon set.

Does labelling hurt your reach?

This is the part where most advice is stale. The old calculus said labelling costs reach, so don't label. Two things inverted it.

Detection labels you anyway. C2PA credentials travel in the file. Platform classifiers catch what credentials miss. TikTok's invisible watermark survives re-encoding. The realistic outcome of not disclosing isn't an unlabelled post — it's a labelled post with a distribution penalty attached.

And the platform's own position is that disclosure is neutral. YouTube's stated position is that disclosing altered or synthetic content doesn't by itself limit your audience or remove monetisation eligibility.

Disclose and keep your reach. Don't disclose, and get labelled anyway.

The honest counter-pressure: some platforms give viewers controls to see less AI content. That's an audience choice rather than a ranking penalty, and it affects disclosed and detected content alike.

Where this falls short. "Doesn't limit your audience" is the platform's stated position, not an independently measured result. Nobody outside the platforms can verify it.

Where is this heading?

From platform policy into statute, and from creator-declared into automatically detected.

Two directions are worth watching. State legislatures are producing AI disclosure obligations faster than Congress is — and marketplaces pick those up as seller requirements, which turns a state law into a practical norm within months rather than years.

Disclosure is a different system from rights, though. Labelling synthetic content isn't the same as having the right to create it.

Platform requirements and statutory obligations described here are current as of the publication date and change frequently. Verify before relying on any of it for a commercial delivery. Nothing here is legal advice.

More on workflow, compliance and delivery in Production.