YouTube Synthetic Content Disclosure: When the Toggle Applies
The Studio toggle is triggered by realism, not by AI. Asset by asset, when YouTube synthetic content disclosure is required and when it is explicitly exempt.
The answer in one sentence: the disclosure question is triggered by how real the content looks, not by whether a generator touched it, which means most AI assisted production work on YouTube requires no disclosure at all and a single photorealistic shot of a place you never filmed does.
That is the opposite of how creators use the setting in practice. The common patterns are both wrong in the same direction: people who declare every video that involved any AI tool, and people who assume the rule is only about deepfaked faces. The rule as written is narrower than the first group thinks and broader than the second.
Everything below was read from YouTube's own disclosure page on 2026-08-26. Platform policy in this area moves faster than anything else in the creator stack, so the date is part of the claim.
Scope, stated up front. This page covers video and audio assets: generated footage, altered footage, synthetic voice, synthetic music, avatars and lip sync. It does not cover disclosure of AI written text, AI text detectors, authorship, or copyright analysis. Those are different questions with different answers and they are not this site's ground.
Where the setting lives
The mechanism is one question during upload. In YouTube Studio, under Attributes, there is an "AI use" setting, and you answer it Yes or No. That is the entire compliance action for this rule. There is no separate form, no submission, and nothing to justify.
What the audience sees depends on what you made. YouTube's page says "a label in the video player may also appear" for photorealistic content, and that "labels may appear in the expanded description for AI content that is non-photorealistic or animated." So the prominence of the disclosure scales with how convincing the content is. An animated piece gets a line in the expanded description. A photorealistic one can get a label on the player itself.
Two consequences of that design are worth noticing. First, the label is not a punishment; it is metadata. Second, declaring Yes on something that did not need it puts a label on your video for no reason, which is a real cost even though it is not a penalty.
The test: would a viewer think this really happened
YouTube's requirement is stated in terms of what the content depicts. Disclosure is required when AI content "makes a real person appear to say or do something they didn't do," "alters footage of a real event," or "generates a realistic scene that didn't actually occur."
Read those three clauses as three separate doors. A real person doing something they did not do. A real event altered. A realistic scene that never happened. Any one of them opens the requirement.
The examples YouTube gives of content that must be disclosed are specific, and one of them surprises most creators: "AI generated music," "AI generated extra footage of a real place," "making it appear as if someone gave advice that they did not actually give," and "depicting a public figure stealing something they did not steal."
AI generated music is on the list. That is the single most commonly missed trigger, because a generated backing track feels like a production choice rather than a depiction. Read at the source, it is named. If you are scoring your videos with a music generator, this rule reaches you even if every frame on screen was shot on your own camera.
"AI generated extra footage of a real place" is the second one people miss. Generating a few seconds of a city you are talking about, to cut between your talking-head shots, is exactly this example. It looks like B roll. Under the rule it is a realistic depiction of a real place that was not filmed there.
What is explicitly exempt
The exemption list is where most working creators actually land, and it is broader than the discourse suggests.
YouTube says creators do not need to disclose "non-realistic content that's made with AI, or edits to realistic content that are minor." Its examples include fantastical imagery, "beauty filters," "color adjustment or lighting filters," and "production assistance, like using generative AI tools to create or improve a video outline."
The production assistance exemption is the important one. Planning, outlining and structuring a video with a generative tool is named on the exempt list. The rule is about what ends up on screen and in the audio, not about how you got there.
The realism exemption does the rest of the work. Something that is visibly not real, a stylized animation, an obviously impossible scene, is not the thing the policy is aimed at, because a viewer is not going to mistake it for a record of an event.
Asset by asset, which way each one falls
This is the table the SERP does not build, and it is the reason to keep this page rather than a summary of it. Every row is an application of the rule above to a specific thing creators actually make. Where a row is a judgment call rather than a named example, it says so.
| The asset | Disclosure question triggered? | Why |
|---|---|---|
| Generated footage of a real place | Yes | Named directly: "AI generated extra footage of a real place" |
| Generated music bed | Yes | Named directly: "AI generated music" |
| A real person made to say something they did not | Yes | Named directly, and the clearest case in the policy |
| A photorealistic scene that never happened | Yes | Named directly |
| An avatar or synthetic presenter delivering your script, presented as a person | Judgment call, and the safe answer is Yes | A realistic human presenter who does not exist reads as a realistic scene that did not occur. YouTube does not name avatars in the examples we read |
| Synthetic voiceover over your own footage | Judgment call | If the voice is presented as a real, identifiable person, the first clause reaches it. A plainly synthetic narrator over real footage is closer to production than to depiction. The policy text we read does not name this case |
| A stylized or obviously animated AI sequence | No | "Non-realistic content that's made with AI" is exempt |
| Beauty filter, color grade, lighting filter | No | Named on the exempt list |
| Outline, script structure or planning done with a generative tool | No | Named as "production assistance" and exempt |
| Upscaling, denoising or minor cleanup of real footage | No, on the face of the rule | "Edits to realistic content that are minor" are exempt. Heavy alteration of footage of a real event is a different matter and falls under "alters footage of a real event" |
Two rows in that table are marked as judgment calls, and that is deliberate. The avatar case and the synthetic narrator case are not among the examples on the page we read. Anyone telling you confidently what YouTube requires for those has gone beyond the published text. When the answer is genuinely unclear, declaring Yes costs you a label and nothing else, which is the cheaper error.
What happens if you get it wrong
YouTube's stated consequence is about a pattern rather than a single video: "creators who consistently choose not to disclose this information may be subject to manual application of a label, or penalties from YouTube, including removal of content or suspension from the YouTube Partner Program."
The first listed outcome is that YouTube applies the label itself. Content removal and Partner Program suspension are named after it, as possible penalties, and the sentence is framed around creators who consistently choose not to disclose.
So an old video you now realize should have been declared is not an emergency. A production line built on undeclared photorealistic synthetic footage is a different risk profile, and this page is not going to reassure you about it.
What this rule is not
It is not the monetization policy. Disclosing correctly does not protect a channel from the inauthentic content rules, and those rules are where channels actually get removed from the Partner Program. They are about originality and are triggered by generic, templated, mass-produced output regardless of whether AI was involved. Different page, different test, different consequence.
It is not a rights clearance. Answering the AI use question says nothing about whether you had the right to depict what you depicted, whether the generator's terms permit your use, or what the law says about the output. Where a rights question comes up, the answer is the vendor's own terms and your own counsel.
And it is not a rule about written content. Whatever you did or did not do with text is outside what this setting asks about, and outside what this site covers.
Read the page yourself
The requirement, the examples and the exemption list live at support.google.com/youtube/answer/14328491. Everything quoted above was read there on 2026-08-26.
That page changes, and the summaries of it drift further from the text with every hop. If a decision matters to you, spend the two minutes on the primary source rather than on anybody's recap of it, including this one.