What Makes an AI Spokesperson Believable

An AI spokesperson is a generated person who presents a brand on video: a digital double of a real employee or model, or a character invented from scratch. The technology is ready. Audiences are wary. Most mistakes happen in the gap between the two.

Two brands, two receptions

In March 2025, H&M announced it would create digital twins of 30 models. The terms came with the news: the models own the rights to their twins, can book them for other brands’ shoots, and get paid each time, like on any campaign. Critics still pushed back. Sara Ziff, of the advocacy group Model Alliance, voiced “serious concerns” about digital replicas without meaningful protections, and said the plan raised critical questions about consent and compensation.

A few months later, a Guess ad in the August 2025 issue of American Vogue featured models generated by an AI marketing agency. No real person stood behind those faces. The reaction on social media was loud, and Condé Nast pointed out that an AI model had never appeared in Vogue’s editorial pages.

Same technology, same year. One project started with the people whose faces were used. The other started with faces that belonged to no one.

What audiences say

Getty Images’ VisualGPS research, based on surveys of 7,500 adults in 25 countries, found that 98% of consumers consider authentic images and videos pivotal to trust, and that almost 90% want to know whether an image was made with AI. It also found that people feel less favorably toward brands that use AI-generated visuals to create people or products.

What consumers say about authentic images

Share of adults who agree, 25 countries

What consumers say about authentic images

Statement	Agree
Authentic images and videos are pivotal to establishing trust	98%
It is important for an image to be authentic	87%
It is getting to the point where I can’t tell if an image is real	76%

Expedia Group found the same reflex in travel: in its 2025 research, fully AI-generated influencers left viewers uneasy, skeptical and annoyed. A synthetic face starts with a trust deficit. The work is to close it.

Test 1: whose face is it?

If the spokesperson is based on a real person, whether a founder, an employee or an actor, you need that person’s written permission, for this use and for a set period. In Quebec, the Civil Code says it plainly: article 36 lists using a person’s name, image, likeness or voice for any purpose other than the legitimate information of the public as an invasion of privacy. The voice is on that list. A cloned voice counts as much as a face.

Plan for the day that person leaves the company. Who controls the double then? Write the answer into the contract before the first frame is generated.

If the character is invented, check that it does not resemble someone real by accident. A generated face that looks like a known actor can create the same problem as a copied one.

Test 2: does the viewer know?

Platforms and lawmakers are converging on disclosure. Since March 2024, YouTube has required creators to disclose realistic altered or synthetic content, including a face swapped onto someone else or a synthetically generated voice. Europe made deepfake disclosure a legal duty in August 2026, and California now requires ads with AI-generated performers to say so.

A label placed early costs nothing and changes the question in the viewer’s head. “Is this real?” becomes “Is this useful?” A good video can win the second one.

Test 3: what can it claim?

This is the hard line. An AI spokesperson can explain, demonstrate and guide. It should never pose as a customer. In the United States, the FTC’s rule on fake reviews, announced in August 2024, bans testimonials that pretend to come from someone who does not exist, and names AI-generated fake reviews as an example. In Canada, the Competition Act already treats publishing a testimonial as reviewable conduct unless the person actually gave it or approved it in writing. A generated customer cannot approve anything.

So a synthetic host presenting your new service in French and English is fine. A synthetic “satisfied client” is a liability.

Where it works, and where it does not

The uses that hold up share a pattern: the viewer wants information more than emotion, the message changes often, and it has to exist in several languages. Onboarding and training videos, product walkthroughs, FAQ clips, monthly updates that would otherwise mean booking a shoot every month. There, a consistent presenter is an asset. Keeping that presenter consistent is a craft of its own.

It struggles with emotional brand stories, apologies, and anything where the viewer needs to believe a human is accountable. Put a real person on camera for those.

Before you approve one

Ask three questions. Whose face and voice is this, and where is the signed permission? Where does the viewer learn that it is AI? Is this character saying anything that only a real customer could honestly say? If all three answers are clear, an AI spokesperson can do real work for a brand. If one is fuzzy, fix it before launch. The audience will ask.