Used AI for a thumbnail, a voiceover or a whole video? Here is a plain-language guide to when YouTube and Instagram expect an AI label, what the label does to reach, and a simple rule to stay safe.
Updated
Open all four studios free — no card, nothing to install.
Almost every creator we talk to now uses AI somewhere: a cleaned-up voice track, a generated background, a thumbnail idea, sometimes the whole clip. And almost all of them ask us the same nervous question — "Do I have to tell people? And will it kill my reach if I do?"
The short answer: disclose when a viewer could be fooled, relax when they couldn't. The longer answer is below. We checked it against each platform's published rules in October 2026; these rules do change, so treat the links at the end as the final word.
Two things happened together. AI video became good enough to pass for real footage, and feeds filled up with low-effort clips made in bulk. Platforms answered with two separate sets of rules, and mixing them up causes most of the confusion:
A label satisfies the first. It does nothing for the second.
Ask yourself: "Could a reasonable viewer think this really happened, or that this person really said or did this?"
If yes, label it. If no, you're usually fine.
| What you made | Label needed? |
|---|---|
| A realistic video of a real person saying something they never said | Yes |
| A voice cloned to sound like a real person | Yes |
| Real footage edited to show an event that didn't happen | Yes |
| A realistic-looking scene of a real place (a flood, a fire, a crowd) that is generated | Yes |
| A fully generated, realistic-looking "person" presenting your video | Yes |
| Cartoon, anime, fantasy or obviously unreal visuals | No |
| Colour correction, filters, background blur, beauty touch-ups | No |
| AI-written script, title, description or automatic captions | No |
| AI-generated background music with no real artist imitated |
The grey area sits in the middle: a product photo with a generated background, for instance. Our own rule in the studio is simple — if we paused to think about it, we label it.
When you upload in YouTube Studio, the details page asks whether the video contains altered or synthetic content. Answer Yes if the table above says so. YouTube then shows a note in the description, and for sensitive subjects — news, elections, health, finance — a more visible label on the video itself.
Three things creators often miss:
Instagram uses an "AI info" label. You switch it on when you post (in the options before sharing, look for the AI label setting). Meta asks you to label photorealistic video and realistic-sounding audio that was generated or altered with AI.
Instagram also reads signals that many AI tools embed inside the file. So an image straight out of a generator may get the label automatically, even if you did nothing. That occasionally catches photos that were only lightly retouched, which is annoying but harmless — the label sits in the post's menu, not across your picture.
India's IT Rules were amended in 2026 to deal with what the rules call synthetically generated information. In simple terms, the direction of travel is: AI-made content that looks real should be clearly labelled, platforms are expected to make labelling possible and check it, and unlawful deepfakes must come down quickly once reported.
We're designers and editors, not lawyers, so take this as orientation rather than legal advice. But the practical upshot is the same as the platform rules: label realistic AI content, and never make a real person appear to say or do something without their consent.
This is the fear behind the question, so let's be direct. Neither platform says a label on its own reduces distribution. A labelled video that people watch to the end and share is treated like any other. What sinks AI content is the same thing that sinks any content — viewers swiping away in two seconds because it looks like a hundred other clips.
Audiences are also getting sharper. A viewer who spots unlabelled AI tends to say so loudly in the comments, and that costs more trust than a small tag ever would.
When we make images and clips in Vision Studio, we treat the output as raw material, not the finished post. The generated picture goes into Pixel Studio for real text, a real logo and a layout; generated clips go into Motion Studio with our own voice and captions on top. By the time it's published, there's enough of us in it that the originality question answers itself — and if the visuals look real, we label them anyway.
If you're new to generating images, start with our prompt formula guide. And if you make AI-assisted Shorts, read what YouTube Shorts rewards in 2026 next.
Platform policies move faster than any blog post. Before a big launch, read the source:
Both YouTube and Instagram say a disclosure label on its own does not limit how a video is recommended. What does hurt is realistic AI content that should have been labelled and wasn't, and content that viewers skip because it feels generic.
The disclosure tools are about the video or post itself. But a thumbnail that shows something realistic that never happened can still break the rules on misleading thumbnails, whether AI made it or not. Keep thumbnails honest about what's in the video.
| No |
| An AI-made illustration used as a clearly decorative graphic | No |
No. Using AI for ideas, outlines, scripts, titles or automatic captions is treated as production help, not as altered content. The label is for what the viewer sees and hears.
A generic AI narrator reading your script usually isn't the concern. A voice cloned to sound like a real person — including a celebrity or someone you know — is, and needs both a label and that person's permission.
Yes, if they're original and add real value. YouTube's partner rules exclude mass-produced, repetitive or near-identical videos, and that applies whether or not AI was involved.
