Labelling AI-generated ads: what to disclose and where

An ad with a synthetic presenter looks like a person recommending a product. Viewers reasonably assume that person exists, which is exactly what advertising rules are built to protect. Labelling is how you keep the format and stay honest about it.

Where disclosure belongs

Treat it as two separate obligations. The first is the platform: the major ad networks ask advertisers to declare AI-generated or digitally altered depictions of real-looking people when uploading a creative, and they can reject or demote an ad that skips the declaration. The second is the creative itself, where a short on-screen line tells the viewer what they are watching without them having to open a settings panel.

• Declare it in the platform field when you upload the ad.
• Keep a visible line in the ad, legible at the size people actually watch.
• Say it in the caption or description too, where the format allows.

Wording that works

Short, plain and specific beats legalese. The label should say that the person is not real, not merely that "AI was used", because AI is used in most editing pipelines today and viewers read that as routine post-production.

• "AI-generated presenter" — clear and short enough for a corner label.
• "Presenter is AI-generated. Product footage is real." — when you mix real product shots with a synthetic person.
• Avoid "enhanced with AI" or "made with technology": both hide the part that matters.

What a label cannot fix

Disclosure addresses who is speaking, not what they say. Consumer protection rules still apply to every claim in the script: results, savings, health effects and comparisons all need support you could show a regulator. A generated script sounds confident by default, which makes this the most common way a well-labelled ad still becomes a problem.

• No testimonial framing for a synthetic presenter: they never bought, tried or recommended anything.
• No implied expertise: a synthetic person in a white coat is a medical claim about a person who does not exist.
• No real person’s face or voice without written permission, whether or not it is labelled.

How AdFace handles it

Every presenter in the library is synthetic by design and described that way in the product, and the acceptable use policy forbids passing one off as a real customer or professional. Scripts pass an automated policy check before they can be approved, which blocks categories such as guaranteed financial returns or medical cures, but that check is a filter, not a legal review: the advertiser remains responsible for the claims and for the disclosure on each platform.