What YouTube's AI disclosure does not cover
YouTube exempts animation, beauty filters, colour work, effects, your cloned voice, captions and audio repair. Decide in one pass if a video needs the label.
Most AI-assisted video on YouTube does not require the altered-content disclosure, and YouTube publishes the exemption list that says so. That list is the useful half of the policy and the half nobody reads, because the headline coverage went to the requirement and not to its boundaries.
The consequence of skipping it is a channel that ticks the disclosure box on everything, defensively, on the theory that over-disclosing is free. It is not quite free. A label that appears on a stylised animation trains your audience to ignore the label, and it signals a production process to competitors and clients that may not match what actually happened. More practically, teams that disclose reflexively usually have not read the rule, which means they also have not noticed which of their videos genuinely does trip it.
Here is the boundary, in one pass.
The three things that trigger it
YouTube's altered or synthetic content disclosure is scoped to content that meaningfully alters or generates photorealistic material, in three named situations:
- Making a real person appear to say or do something they did not. Synthetic speech in someone's voice, a face swapped onto footage, a performance they never gave.
- Altering footage of a real event or place. Changing what a real location or a real incident looked like, in footage presented as depicting it.
- Generating a realistic scene that never occurred. A photoreal depiction of an event that did not happen, presented as though it could have.
Two words carry the whole rule. Photorealistic, which is why stylised output falls outside it. And meaningfully, which is why incidental processing falls outside it. If a video is neither photorealistic nor meaningfully altered in one of those three ways, the disclosure requirement does not attach.
The published exemption list
YouTube names these as not requiring disclosure. It also states that the list is not exhaustive.
| Exempt | What it covers |
|---|---|
| Animation and clearly unrealistic content | Fantasy, stylised, obviously non-photoreal output |
| Beauty filters | Skin, face and appearance filters |
| Colour adjustment | Grading, correction, look development |
| Special effects | Background blur, vintage looks, ordinary effects filters |
| Cloning your own voice | Your voice, used by you |
| Caption generation | Automatic captions and subtitles |
| Audio repair, sharpening, upscaling | Cleanup, restoration, enlarging footage |
| Production assistance | Outlines, scripts, thumbnails, titles, infographics, idea generation |
| Gameplay footage | Video captured from a game |
Read that list against a typical AI-assisted upload and most of the pipeline is on it. A generated script, an automatically captioned edit, a graded export and a synthetic voiceover in your own cloned voice is four exempt items and no trigger.
The other column matters too. YouTube's own examples of content that does require disclosure include making a real person appear to give advice they did not give, a realistic weather event that did not happen, extra photoreal footage of a real place, and AI generated music. Do not fold a generated score into "production assistance". That exemption is scripts, titles, thumbnails, outlines and ideas, not a generated soundtrack.
Three of the exemptions are worth expanding because they are where teams misread the boundary.
Cloning your own voice is exempt. Cloning someone else's is trigger one. The distinction is whose voice, not whether cloning happened. A narrator using a model of their own voice across a catalog is doing something the policy explicitly exempts. The same technique applied to a public figure's voice is the first trigger case, and it also raises rights questions the disclosure policy does not address at all. If voice is central to your production, the AI voice cloning tool is where that decision lives, and consent for the voice is a separate matter from disclosure to YouTube.
Animation is exempt because of how it looks, not how it was made. A fully generated animated short is exempt on the same basis as a hand-drawn one, because the exemption is about photorealism. Conversely, a scene made with entirely conventional tools can be photorealistic and can trip the rule. The technique is not what is being regulated.
Production assistance means the invisible half of the pipeline. Scripts, titles, ideas and research assistance are named as exempt. This matters for anyone running an agent-driven workflow where the model is doing planning and drafting rather than generating pixels. Nothing about that half of the process creates a disclosure obligation.
Deciding in one pass
Four questions, in order. Stop at the first yes.
- Does the video contain photorealistic footage of a real, identifiable person doing or saying something they did not? → Disclose.
- Does it alter footage of a real event or place in a way that changes what the viewer would understand happened there? → Disclose.
- Does it present a realistic scene of something that did not occur, in a way a viewer could take as real? → Disclose.
- Is everything else you did on the exemption list? → No disclosure required.
If a video reaches question four and something is not clearly on the list, that thing is worth naming specifically rather than defaulting either way. The realistic ambiguous cases are: a photoreal presenter who is not a real person, a generated environment standing in for a real location, archival-looking footage of a historical event, and a generated soundtrack. YouTube lists AI generated music among the examples that require disclosure, so do not wave a generated score through as production assistance. The other three sit close enough to the visual triggers that the safer read is to disclose.
That routine is the YouTube column of a wider matrix. The cross-platform version, including where each other platform draws its own line, is in one disclosure, five destinations.
The label is not the risk
This is the part that changes how you should feel about the whole question. YouTube's own disclosure page states that the disclosure does not limit audience reach or monetization eligibility.
Take that at face value: the label itself does not cost you distribution or revenue. Disclosing where the rule applies is not a performance decision that needs weighing against a compliance one. It is just the rule, and applying it is close to costless.
The actual monetization surface sits elsewhere entirely. In July 2025 YouTube renamed its "repetitious content" policy to inauthentic content, targeting mass-produced and templated output that is easily replicable at scale. That policy is indifferent to whether AI was involved. A channel producing visibly templated video at volume can fail it with no generated frames at all, and a channel producing genuinely distinct video can pass it while disclosing on every upload.
So the correct mental model is two independent tests. Disclosure asks what does this depict. The inauthentic content policy asks is this shelf of videos meaningfully distinct from each other. Failing the second is what actually threatens a channel. What YouTube counts as mass-produced video breaks down the second test, and surviving a YouTube inauthentic content strike covers what happens when it goes wrong.
The consequences YouTube names for not disclosing where you should have are worth knowing regardless: YouTube may apply the label itself, remove the content, or suspend the channel from the Partner Programme. Given that disclosure is stated to cost nothing in reach or revenue, that risk profile is entirely one-sided.
Where this lands in a pipeline
Three placements, in increasing order of usefulness.
At the brief. You know at brief stage whether a video will depict a real person, a real place or a realistic event. That is the whole test, and it is answerable before anything is generated. Recording the answer on the brief means the upload step is reading a decision rather than making one.
At the export. YouTube also auto-labels based on C2PA provenance metadata and on content made with its own generative tools, so in some cases a label arrives without you selecting anything. Whether that metadata survives your pipeline is a separate and surprisingly fragile question, covered in why C2PA manifests die on upload. Do not rely on automatic labelling to satisfy a requirement you are obliged to declare yourself.
At the upload. Select the disclosure where your brief says to. Do not treat it as a judgement call made by whoever happens to be posting, because that is how a channel ends up with inconsistent answers on similar videos, which is a worse look than either consistent answer.
If monetization thresholds are the reason this matters to you, YouTube monetization thresholds for AI channels covers the entry requirements, which are a separate gate again from either of the two policies above.
FAQ
Does using an AI video model at all require disclosure?
No. The requirement is scoped to photorealistic content in three specific situations, not to the use of a tool. A generated animation, a stylised sequence or a clearly unrealistic scene falls outside it regardless of how it was produced. Other platforms draw the line in their own places, which is why a single house rule across all of them tends to be either too strict or too loose.
If I clone my own voice, do I need to disclose?
YouTube names cloning your own voice as exempt from this disclosure requirement. That is specific to your own voice. Using a model of someone else's voice is the first trigger case, and it also raises consent and likeness questions that are separate from anything YouTube's disclosure policy addresses.
Does disclosing reduce views or demonetize a video?
YouTube's own page on the requirement states that it does not limit audience reach or monetization eligibility. The demonetization risk for AI-heavy channels comes from the inauthentic content policy, which is a different rule and does not care whether you disclosed.
What happens if I should have disclosed and did not?
YouTube states it may apply the label itself, remove the content, or suspend the channel from the Partner Programme. Since disclosure is stated to carry no reach or revenue cost, there is no upside to omitting it where the rule applies. Treat borderline cases as synthetic media disclosure cases and label them.