Say What’s Synthetic

The short answer. If AI made the voice, the face or the footage in a video, I’d put one plain sentence saying so in the description, and I wouldn’t use a synthetic person as a testimonial. YouTube, Meta, the FTC and the EU Commission each say something about this, and what they say is below. I’m reporting what the pages say. I’m not telling you what you must do.

Picture a breathing practice. A slow voice, an even pace, a pause before each count. Under it, a comment: “Your voice is so steady. Where do you teach?”

The voice is software. You wrote the script, chose the voice and pressed export. The comment is kind, and it’s about someone who isn’t there.

You didn’t set out to mislead anyone. Nobody asked, and you didn’t say. That small gap is what this piece is about.

What the platforms and regulators actually say

I read the primary pages. These are the parts that matter here, in their words or close to them.

YouTube. Its help page on disclosing generative AI says YouTube requires creators to disclose GenAI content that makes a real person appear to say or do something they didn’t, alters footage of a real event or place, or generates a realistic scene. It says non-realistic content and minor edits don’t need the disclosure, and gives examples such as “Someone riding a unicorn through a fantastical world.” (YouTube Help: Disclosing use of GenAI content) Whether your video falls on one side of that line is a question for that page, not for me.

Meta. In 2024 Meta said the “AI info” label goes on when it detects industry-standard signals that content is AI-generated, or when the person posting says so themselves. (Meta: approach to labeling AI-generated content, April 2024, updated through 12 September 2024) So there are two routes to the label. One is that the file carries the signals. The other is that you tell people.

The FTC. On 14 August 2024 the US Federal Trade Commission announced a final rule banning fake reviews and testimonials. The release says the rule covers testimonials by “someone who does not exist, such as AI-generated fake reviews.” (FTC press release)

The EU Commission. Its page on transparency rules for AI systems says the rules apply “from 2 August 2026” and that “Deployers must disclose deepfakes.” The page describes a deepfake as content that “resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful.” It also describes a grace period until December 2026 for one provider duty, marking AI-generated content, for generative systems already on the market before 2 August 2026. That is a duty on the makers of the systems. Read the page before you decide what applies to you. (European Commission: Quick facts, transparency rules for AI systems)

None of this is legal advice.

The sentence

Here is the sentence:

The narration in this video is an AI-generated voice.

It’s one sentence, with no apology and no “disclaimer:” in front of it. It names the thing, narration, and says what it is. Someone reading the top of the description can know what they’re hearing before they press play.

If the thing that’s synthetic is something else, change the noun. The images in this video are AI-generated. The music in this video is AI-generated. Keep it as short as that, and say it about the thing that’s actually synthetic.

Put it in the description, in plain text.

YouTube has a setting for AI content, and it’s worth looking for. But a setting lives in a menu. The sentence is right there in the text, where a viewer can read it without opening anything.

Here is where I am with my own. On the Scribe Hands page, the narrated explainer carries that exact line directly under the player. On YouTube, five of my videos say in their descriptions that the voice is synthetic, mostly in the same words (“The narration is a synthetic voice.”). The films I’m staging carry the sentence as written.

The one thing I wouldn’t synthesize

A testimonial is a person saying it worked for them. If the person doesn’t exist, it didn’t work for anyone.

That’s the mechanism, and it’s why the FTC line stuck with me. An AI-written voiceover is a production choice, and you can tell people about it in one sentence. A made-up person saying “this changed my practice” is a claim about something that never happened. I don’t think a sentence at the top of the description fixes that, so I don’t make them.

If you have a real student who will say it in their own words, that’s a testimonial. If you don’t yet, an empty slot is better than a full one.

What this doesn’t settle

Whether AI belongs in wellness content at all, whether it helps people or harms them, and what it does to anyone’s wellbeing. I haven’t seen a source on those I’d trust, so I’m not saying anything about them. This piece is narrower. It’s about one sentence, and about not putting a person on screen who isn’t one.

Questions people ask

Do I have to label an AI voiceover on YouTube?
I can’t tell you that. The page says YouTube requires disclosure of realistic altered or synthetic content in three cases, and says non-realistic content and minor edits don’t need it. Its list of what doesn’t need disclosure also includes “Cloning one’s own voice to create voice overs or dubs.” Read it against your own video.

What does Meta’s “AI info” label do?
Per Meta’s post, it’s applied when industry-standard signals are detected or when the person posting discloses it. If you disclose, the label doesn’t depend on whether anything detected the file.

Can I use an AI-generated person as a testimonial?
The FTC’s 2024 release says its rule covers testimonials by someone who does not exist. That’s the page to read.

Your next step

Write your own line, in one sentence, before your next upload. Name what’s synthetic and say what it is. Put it in the description.

Say it before anyone asks. A question you answer first reads as a note. The same answer given after reads as a correction.

thequietai.com