Where avatars work
Avatars earn their place where the message matters more than the messenger and changes often: weekly internal updates, onboarding, compliance and product training, help-centre walkthroughs, and the same explainer in many languages. In these uses, audiences want clarity and consistency, and re-rendering a changed line saves a reshoot. A presenter who speaks the viewer's language, with timing matched, can be more useful than subtitles under a foreign speaker. Keep delivery plain, backgrounds simple and scripts short: avatars are least convincing in long, emotional takes and most convincing in clear, practical ones.
Where they backfire
Avatars go wrong where trust depends on a real person being present. Brand advertising is one case: audiences who spot a synthetic face in an ad may read it as cheap or evasive, and it raises the question of what else is not real. Sensitive subjects are another: redundancies, safety incidents, health results, condolences, apologies and crisis statements should come from an accountable human. Avoid avatars for anything that looks like personal advice in regulated areas such as finance or health, and never script an avatar of a real executive to say anything that person has not approved word for word.
Consent and contracts for a real person's likeness
A presenter cloned from a real employee, actor or executive needs a contract written for that purpose, not a standard image release. It should cover the scope of use, media, languages and markets, duration, payment, the person's right to review scripts, what happens if they leave, and deletion of the face and voice model at the end. Face and voice recordings are personal data under the GDPR. In the US, likeness and voice are protected by state law; California's AB 2602, in force since January 2025, can make a digital-replica clause unenforceable when the replica does work the person would otherwise have done in person, the uses are not reasonably specifically described and the person had no lawyer or union representing them.
Disclosure under the EU AI Act
Article 50(4) of the EU AI Act (Regulation (EU) 2024/1689) requires anyone deploying an AI system to create a deepfake to disclose that the content was artificially generated or manipulated. A deepfake here means realistic image, audio or video that resembles existing people, places or events and would falsely appear authentic, and the Commission's guidance reads this as including people who could plausibly exist. The duty applies from 2 August 2026. A photorealistic avatar, whether of a real executive or an invented presenter, is in scope; a clearly stylised cartoon presenter is unlikely to count. Put the disclosure in the video itself, not only in the description, and add C2PA content credentials where platforms read them. YouTube separately requires creators to label realistic altered or synthetic content.
Updated 25 September 2026