This is the highest-risk format in synthetic advertising, because the entire persuasive weight rests on a viewer believing a person exists and holds an opinion. Nothing else in generative production makes so specific a claim about reality.
Both variants carry problems and they are different problems. An invented endorser is a misleading-advertising question. A real person’s likeness endorsing something they did not endorse is a likeness question, a defamation question and now an Article 50 question at once.
Our position is simple and has never cost us a job worth having: a synthetic presenter may not make a claim the brand could not make in its own voice, and the format is disclosed on the asset.
Can a synthetic presenter give a product testimonial?
Only with disclosure, and only saying things the brand could substantiate in its own voice. The persuasive power of the format comes from implied independence, which is exactly what makes an undisclosed version misleading.
Is an invented endorser safer than using a real person’s likeness?
It removes the likeness problem and keeps the misleading-advertising problem. Both need disclosure; the second also needs an explicit release.