Will customers know it’s a 数字人? And should you tell them?

Will they actually notice?

On a fast scroll, some won’t. The ones who stop and watch usually do. The giveaways are small, but they add up: the delivery is a little too even, the hands and shoulders hardly move, the mouth is slightly off on certain sounds, and the same face says the same kind of sentence every week. Avatars get better every few months. But the people most likely to notice are your regulars, your existing clients and your competitors, which is exactly the group whose trust you care about.

Noticing isn’t the damage. Nobody unfollows a bakery because its holiday-hours video is an avatar. The damage comes when someone notices and feels they weren’t supposed to.

The platforms are labelling it anyway

Even if your avatar fooled everyone, the choice is increasingly not yours. The main platforms now label realistic AI video:

• TikTok expects creators to switch on its AI-generated content label for realistic AI people and scenes. It labels content automatically when the file carries content credentials from the tool that made it, and ads have their own AI disclosure.
• Facebook and Instagram add an “AI info” label when they detect industry-standard AI signals or when you disclose it yourself, and Meta asks people to disclose photorealistic video or realistic audio that was digitally created.
• 小红书 and other mainland platforms work under China’s AI content-labelling rules, in force since 1 September 2025. Those rules require AI-generated content to be marked, and the duty sits with publishers as well as platforms.

Platform rules change often, so check each one’s current help page before you post. But they only move in one direction. Assume your 数字人 will carry a label whether you add it or not, and remember that a label you added reads very differently from one that was added for you.

The rule that actually bites: don’t mislead

Singapore has mostly governed AI through guidance and frameworks rather than one sweeping AI law. That doesn’t mean anything goes. The rule that matters for a small business is an old one: don’t mislead customers. An avatar presented as a real customer, a real review, or a live person replying in real time turns a marketing shortcut into a misleading claim.

Using someone’s face and voice is its own question. Get the person’s agreement in writing, including where the avatar will be used and what happens to it if they leave.

How to disclose without killing the post

Disclosure doesn’t have to read like a disclaimer. Done well it’s one line, and it can even work in your favour, because it shows you’re organised and up to date.

• Switch on the platform’s AI label for every avatar post. It costs nothing and takes away the “gotcha”.
• Say it once in the caption, plainly. In English: “AI avatar of our host, script by our team.” In Chinese: “本视频由数字人出镜，内容由我们团队撰写。” Put it at the end, not the start.
• Keep the same person real elsewhere. When the face in the avatar also turns up live every week, the avatar reads as a convenience, not a disguise.
• Pin one explainer. A short post from the real person (“some of our updates use an AI version of me, and here’s why”) answers the question once for everyone who wonders.
• If someone asks “is this real?”, tell the truth. A straight “yes, that’s our AI avatar; the real one does the Wednesday videos” wins the comment thread.

Will the label cost you views? Some posts will dip and some won’t. What decides reach is still whether the post is useful, and an honest, useful avatar post beats a hidden, generic one.

Where a 数字人 should never stand in

Some posts only work because a human made them. Keep the avatar out of these:

• Testimonials and reviews. An avatar “customer” is a fake testimonial. Full stop.
• Apologies and bad news. A synthetic face saying sorry makes it worse.
• Anything live. Live sessions, real-time replies, “I just saw your message”. If the post implies a person is there right now, a person has to be there.
• Personal stories and milestones. The point of these posts is that someone cared enough to say it.
• Anyone who hasn’t agreed. No avatar of a staff member, client or partner without their written consent.

How we run it ourselves

Our own Instagram and Facebook use a 数字人 of the same team member who appears on camera for real, and we hold ourselves to the rules above: label the avatar clips, keep the real-person videos running alongside them, and cover the likeness with a short written agreement. Our branding and social media service is built on the same rules, so clients get the time saved without the trust problem. This is part four of our 数字人 series. If you’re new to it, start with what a 数字人 is.

Common questions

Do I have to label AI avatar videos?

Platforms increasingly require it. TikTok asks for an AI label on realistic AI people, Meta labels or asks you to disclose photorealistic AI video, and mainland platforms such as 小红书 follow China's labelling rules. Separately, never present an avatar in a way that misleads customers.

Will labelling my 数字人 reduce engagement?

Some posts may dip, but usefulness still decides reach. A labelled, useful post beats a hidden, generic one, and labelling it yourself avoids the worse outcome of being caught.

Can I use a 数字人 for customer testimonials?

No. An avatar presented as a customer is a fake testimonial. Use real customers with their permission, or don't use testimonials.

Can I make a 数字人 of a staff member?

Only with their clear written agreement, covering where the avatar will be used and what happens to the likeness if they leave.

What should the disclosure say?

One plain line at the end of the caption, such as 'AI avatar of our host, script by our team', plus the platform's own AI label.

The 数字人 series: what a 数字人 is, should your SME use one, real you vs 数字人, with a sample month, and our branding & marketing service.