// HOW-TO · PLATFORM POLICY

How to disclose and document AI content on YouTube (2026)

Disclose and document AI content on YouTube (2026): apply the altered-content label, enroll in likeness detection, and keep the consent records a claim needs.

Last verified · 2026-08-22 · by Moe Ameen

YouTube's 2026 rework of how it handles AI didn't just change what happens after a claim lands — it made two upstream habits load-bearing. First, you have to disclose realistic AI when you upload. Second, once likeness detection and the renamed Claims menu exist, you have to be able to produce, on demand, proof of who is in a synthetic video and that they agreed to it. Get both right before you hit publish and the whole claims process becomes a non-event.

This is the proactive workflow — the disclosure and provenance side, not the dispute side. If a claim has already landed on one of your videos, the response steps live in the companion tutorial on [how to respond to a YouTube AI likeness claim](/how-to/respond-to-a-youtube-ai-likeness-claim); for the full policy background on both systems, see the guide on [YouTube's AI claims process](/guides/youtube-ai-claims-process). Here the goal is simpler: never trigger the claim you can't clear, and always have the paperwork if one comes anyway.

The steps

Run the "realistic depiction" test before you upload. Disclosure is not required for everything AI touched — it is required when your video realistically depicts something that did not happen. That means a synthetic version of a real person's face or voice, altered footage of real events or places, or a realistic scene of an event that never occurred. It is not required for clearly unreal animation, beauty or color filters, or using a model to draft your script or captions. Decide which bucket each upload falls into before you open Studio, because the answer sets everything that follows.
Apply the altered-or-synthetic-content label in YouTube Studio. When a video needs disclosure, add it during upload: in YouTube Studio, at the video-details step, open the "Altered or synthetic content" section and select "Yes." YouTube then surfaces a label to viewers — below the player and above the description on long-form, as an overlay on Shorts — with an expanded on-player label for sensitive topics like health, news, elections, and finance. You can also add or edit the disclosure on already-published videos from the same details page.
Know how auto-labeling works so it never surprises you. Since May 2026, YouTube's systems apply their own signals to detect significant photorealistic AI, and can label a video automatically when the creator didn't. Disclosing yourself is still the better move: you control the framing and timing rather than getting a label stamped on after the fact, and you avoid the read that you tried to hide it. Treat self-disclosure as the default for any realistic synthetic upload, not a step you weigh each time.
Enroll in likeness detection to protect your own face. Likeness detection is the Content ID-style scanner for your face — enroll and YouTube runs a one-time scan of new uploads across the platform, alerting you to videos that may depict you. Set it up in YouTube Studio under the Content Detection section's "Likeness" tab; you must be in the Partner Program, over 18, and complete verification with a government ID and a short face video. It is opt-in and currently visual only (audio is planned), so enrolling protects you — it does not scan the platform for anyone who hasn't signed up.
Capture consent for every real person you depict. The Explicit Consent dispute lane asks for verified permission to use a person's voice or visual likeness, and "they said it was fine" does not clear it. Before you publish anything depicting a real, identifiable person — talent, a client, a collaborator — get a signed release, license, or written agreement that names the use. Capture it at production time, not after a claim arrives, so consent is a record you can produce rather than one you have to reconstruct.
Keep a per-video provenance file. For each synthetic upload, keep one place that answers the questions a claim will raise: what tool generated it, whose face and voice it uses, the source footage or project files, the disclosure you applied, and the consent document if a real person appears. Camera metadata and original project files are also what clear the "not made with AI" lane if a real video gets false-flagged. A tidy record turns a dispute into a lookup instead of a scramble.
Turn it into a repeatable pre-publish checklist. Bundle the above into three questions you run on every upload: Is this a realistic depiction of something that didn't happen? — if yes, toggle the disclosure. Does a real person appear? — if yes, confirm consent is on file. Do I own the face and voice? — if it's your own persona, you're already in the "content I control" lane. Running the checklist by habit is what keeps the claims process a formality rather than a fire drill.

Common gotchas

Disclosing AI is not a penalty; failing to disclose is the risk. The label itself does not suppress reach or demonetize — but skipping a required disclosure on realistic synthetic content is what invites enforcement and an auto-applied label you didn't control.
Over-disclosing muddies the signal too. AI-assisted scripting, captions, obvious animation, and beauty filters do not need the label; toggling "Yes" on those trains viewers to ignore a label that's supposed to mean 'this realistic thing isn't real.'
Likeness detection is opt-in and visual-only. Enrolling protects your face; it does not automatically find deepfakes of people who never signed up, and it does not yet match voices. Don't assume the platform is scanning on your behalf unless you enrolled.
Informal permission is not consent evidence. The Explicit Consent lane needs a verifiable record — a release, license, or written agreement captured before publishing — not a screenshot of someone saying 'sure.'
Deleting and re-uploading to dodge a label re-triggers the scans. You lose the video's history and land back in the same detection pass — fix the disclosure on the existing upload instead.
Auto-labeling can apply a disclosure you didn't choose. If YouTube's systems detect photorealistic AI you left undisclosed, they can label it for you, on their wording and timing — self-disclose to keep control.

Legal note

AI-likeness and publicity rights are jurisdiction-specific and evolving (right of publicity, privacy, and emerging deepfake and 'digital replica' statutes). YouTube's disclosure toggle and likeness detection are platform mechanisms, not legal rulings — applying a label or clearing a claim doesn't settle underlying rights. If you depict a real person, especially a public figure, licensed talent, or any commercial use, obtain and retain documented consent and consult a qualified attorney.

Where Kompozy fits

The two hard steps here are decisions and paperwork: which uploads actually need the disclosure toggle, and being able to prove whose face is in a synthetic one. Both get easier when your synthetic video comes from a single engine that already knows the answer. [Kompozy](/) is a full AI content generation and multi-platform publishing engine, and its output formats are typed — an avatar format like [Persona Shorts](/glossary/persona-shorts), Persona HeyGen, or [Persona Frames](/glossary/persona-frames) is a realistic depiction that gets the disclosure, while a [Text Post](/glossary/output-buckets) or Blog Article drafted with AI is production assistance that does not. So the step-one 'realistic depiction test' stops being a judgment call per video and becomes a property of the format you chose.

The provenance step is where an engine beats a scattered tool-chain outright. Every avatar render in Kompozy descends from your own AI Influencer [persona pool](/glossary/persona-brief) — a face and voice you set up and control — so the 'whose likeness is this?' question has a permanent, documentable answer: yours. That also flips likeness-detection enrollment in your favor. Because you never build on a scraped public figure, enrolling protects your identity against other people's deepfakes rather than exposing you to a claim you'd have to fight. One [Persona Brief](/glossary/persona-brief) keeps that identity consistent whether a piece lands as a Short, a carousel, a blog, or an email, so a month of output shares one consent basis instead of a dozen loose ones.

Kompozy will not manufacture consent for someone else's face or decide the law for you — owning what you depict is your call. What it removes is the per-upload overhead of proving it: typed formats make the disclose-or-not decision deterministic, an owned persona makes provenance a lookup, and generation stays inside one system you can point to when a claim asks who's in the video. Creator ($49/mo for 2,500 credits) fits a solo creator running one owned persona across platforms; Pro ($299/mo for 18,000 credits) suits an agency or brand governing several personas and high upload volume; Enterprise is custom.

Frequently asked questions

How do I add the AI disclosure label on YouTube?

During upload in YouTube Studio, at the video-details step, open the "Altered or synthetic content" section and select "Yes." YouTube then shows viewers a label — below the player on long-form videos, as an overlay on Shorts, and an expanded on-player label for sensitive topics. You can also add or edit the disclosure on an already-published video from the same details page.

What AI content actually requires disclosure on YouTube?

Content that realistically depicts something that didn't happen: a synthetic version of a real person's face or voice, altered footage of real events or places, or a realistic scene of an event that never occurred. It is not required for clearly unreal animation, beauty and color filters, or using AI to draft a script or generate captions — that's production assistance, not a realistic fake.

How do I enroll in YouTube likeness detection?

In YouTube Studio, go to the Content Detection section and open the "Likeness" tab. You must be in the YouTube Partner Program, over 18, and complete verification with a government-issued ID and a short face video. Once enrolled, YouTube runs a one-time scan of new uploads and alerts you to videos that may depict your face. It's opt-in and currently visual only.

What do I need to document to prevent a claim I can't clear?

Proof of consent for any real person you depict — a signed release, license, or written agreement captured before publishing — plus a per-video record of what generated the video and whose likeness it uses. Keep original footage and project files too; those clear the 'not made with AI' lane if a genuine video gets false-flagged. The cleanest prevention is to depict only a persona you own.

Does disclosing AI content hurt my reach or monetization?

YouTube has framed the disclosure label as an informational transparency tool, not a penalty — the label alone doesn't suppress distribution or demonetize. The real risk runs the other way: failing to disclose realistic synthetic content invites enforcement and an auto-applied label you didn't control. Disclose when required and monetization is governed by the usual originality and policy rules, not the label.