This article is general information, not legal advice. Where a term has a legal definition, that definition usually applies only in the place whose law defines it — the same word can mean different things in different countries, and in a contract it means whatever the contract says it means. Definitions here are quoted or paraphrased from primary sources, with links at the end.
The vocabulary around AI likeness is a mess. Vendors, platforms, regulators and legislators use overlapping words for overlapping ideas, and contracts routinely use a term without defining it. This glossary is the one we use across the rest of the blog. Where a definition is lifted from a law or a standard, we say which one.
A practical warning before you start: in an agreement, never rely on the general meaning of a term. If a contract grants rights to your “likeness”, ask whether that includes your voice, your name and your mannerisms, because in several of the definitions below it does not.
The identity terms
Likeness. In ordinary use, anything that makes you recognisable. Legal definitions are narrower and oddly specific — Washington State, for example, defines likeness as a representation “other than a photograph, of an individual’s face, body, or parts thereof, or the distinctive appearance, gestures, or mannerisms of an individual” (RCW 63.60.020). Note that “gestures or mannerisms” are in there, and that photographs are handled separately.
Digital replica. The term the newer US instruments use. The federal bill as reported defines it as “a newly created, computer-generated, highly realistic electronic representation that is readily identifiable as the voice or visual likeness of an individual” in content they did not actually perform, or where their real performance was “materially altered” (S. 4591). See digital replica laws in the US.
Forged digital likeness. Washington’s term for a persistent or real-time visual or audio representation of an identifiable person that has been altered “to be indistinguishable from a genuine” recording, “misrepresents the appearance, speech, or conduct of the individual”, and “is likely to deceive a reasonable person” (RCW 63.60.020). The three-part test matters: alteration alone is not enough without misrepresentation and likely deception.
Personality / right of publicity / personality rights. The commercial right in your identity. Washington defines a “personality” as any individual whose “name, voice, signature, photograph, or likeness has commercial value, whether or not” they commercially exploit it themselves (RCW 63.60.020). Terminology varies by country, and some countries have no such right at all, protecting identity through passing off, privacy or consumer law instead.
Voice clone. A model trained on recordings of a real voice, able to generate new speech in it. Legally it is often treated alongside visual likeness, but practically it behaves differently — it can produce output instantly, in bulk, and often with no screen to carry a label. See voice cloning and consent.
Biometric data. Under the EU AI Act, “personal data resulting from specific technical processing relating to the physical, physiological or behavioural characteristics of a natural person, such as facial images or dactyloscopic data” (Article 3(34)). The UK regulator stresses the same nuance: it is the specific technical processing for unique identification that makes data biometric, so a recording on its own usually is not (ICO).
The AI terms
AI system. The EU AI Act’s definition: “a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions” (Article 3(1)).
Synthetic media / synthetic content. Image, audio, video or text generated or substantially modified by an AI system rather than captured. A useful umbrella term with no single legal definition; the duties attach to narrower categories like deep fakes and marked synthetic output.
Deep fake. Narrower than most people assume. The EU AI Act defines it as “AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful” (Article 3(60)). The test is realism and apparent authenticity — which means a photorealistic presenter depicting nobody real can still fall inside it.
Generative AI. Everyday term for systems that produce content. Not an EU AI Act category; the Act works through AI systems, general-purpose AI models, and the obligations attached to particular uses.
AI avatar. A synthetic presenter, usually from a tool’s library, generally not intended to resemble a specific real person. Whether it resembles one anyway is a question to ask the vendor — see avatars, licensed creators or a video shoot.
Provider and deployer. The EU AI Act splits duties between them. A provider “develops an AI system or a general-purpose AI model” and places it on the market under its own name or trademark; a deployer is anyone “using an AI system under its authority” outside personal non-professional activity (Article 3(3), 3(4)). If you run ads with a bought-in tool, you are usually the deployer, and the disclosure duty for deep fakes sits with you — see our Article 50 guide.
The provenance terms
Provenance. The history of a file: what it was made from and what happened to it. C2PA describes it as understanding “the history of an asset as represented by the set of C2PA Manifests for an asset and its ingredients” (C2PA 2.4 explainer).
C2PA. The Coalition for Content Provenance and Authenticity, the standards body behind the provenance specification (c2pa.org).
Content Credentials. “The preferred non-technical term for a C2PA Manifest” — the user-facing name for provenance data attached to a file (C2PA 2.4 explainer). See what are content credentials?
Manifest and assertion. A manifest is “the set of information about the provenance of an asset consisting of one or more assertions that are digitally signed”; an assertion is “a data structure which represents a statement about the asset” (C2PA 2.4 explainer).
Machine-readable marking. Metadata or a watermark a machine can read, as distinct from a visible label a person reads. The EU AI Act requires both kinds in different situations.
AI label / disclosure. The visible or audible notice that content is AI-generated. Each platform has its own name and its own switch — TikTok’s AIGC label, YouTube’s “AI use” setting, Meta’s AI-disclosure tool. Our platform-by-platform guide maps them.
The consent and contract terms
Consent. In data protection law, a high bar: “any freely given, specific, informed and unambiguous indication of the data subject’s wishes by which he or she, by a statement or by a clear affirmative action, signifies agreement” (ICO). Silence, pre-ticked boxes and default settings do not qualify, and it must be withdrawable without detriment. See consent, not deepfakes.
Scope. What a licence actually permits: uses, media, territories, dates, categories. Vagueness here is now a legal weakness as well as a bad deal — New York voids a digital replica provision lacking “a reasonably specific description of the intended use” where the person was unrepresented (S7676B), and California took a comparable approach (AB 2602).
Term. How long the licence lasts. The federal bill as reported would cap a lifetime licence at “10 years”, and at “5 years” for a minor, ending at 18 (S. 4591).
Exclusivity. Whether you may license to others — nobody, nobody in a category, or anyone. The main lever in any likeness negotiation; see pricing a likeness licence.
Buyout. A single payment for unlimited, usually perpetual use. Not a licence in any meaningful sense, and the thing most likeness advice tells creators to refuse.
Model release. The conventional consent document from photography and film, covering footage of a real person. Most were written before AI replicas existed and should not be assumed to cover them.
Right to erasure. Under UK and EU data protection law, the right “to have personal data erased”, also called the right to be forgotten, with a response due “without undue delay and at the latest within one month” (ICO). In a likeness context this is what you lean on to get recordings and trained models deleted — but say so explicitly in the agreement too, with a deadline.
The advertising terms
Endorsement. A message consumers would understand as reflecting the opinions or experience of someone other than the advertiser. The FTC’s position on the limit is short: “you can’t talk about your experience with a product if you haven’t tried it” (FTC).
Testimonial. An endorsement based on personal experience. The FTC’s 2024 rule prohibits testimonials that misrepresent that they come from “someone who does not exist, such as AI-generated fake reviews”, or from people without genuine experience of the product (FTC).
Material connection. Any relationship between an endorser and a brand that consumers would not expect and that would affect how they weigh the endorsement. It has to be disclosed clearly and conspicuously (FTC).
Ad recognition. The requirement that advertising be identifiable as advertising. The UK code puts it as “marketing communications must be obviously identifiable as such”, and bars a marketer from implying it is “acting as a consumer” (ASA). The reason AI UGC ads need care.
Substantiation. Holding evidence for a claim before you make it. Who or what delivers the claim on screen changes nothing about this.
Where Likeness fits in
Likeness is being built to respect these distinctions rather than blur them: voice and appearance as separate grants, scope stated in specifics, a term with an end date, consent that can be withdrawn, and labelled output tied to a licence record. It is in development and not yet open — see for businesses, for creators and trust and consent, or email [email protected].
General information only, not legal advice, and no claim that any definition here is the operative one in your country. Laws, standards and platform policies in this area change often — check the current primary sources for your markets.
Sources
- Washington State Legislature, RCW 63.60.020: Definitions
- US Government Publishing Office, S. 4591 bill text as reported (NO FAKES Act of 2026)
- New York State Senate, S7676B
- California Legislature, AB 2602 (2024)
- European Commission AI Act Service Desk, Article 3: Definitions
- European Commission AI Act Service Desk, Article 50: Transparency obligations
- C2PA, Content Credentials specification 2.4 explainer
- C2PA
- Information Commissioner’s Office (UK), What is valid consent?
- Information Commissioner’s Office (UK), Right to erasure
- Information Commissioner’s Office (UK), Biometric data guidance: key data protection concepts
- Federal Trade Commission, FTC’s Endorsement Guides: What People Are Asking
- Federal Trade Commission, FTC announces final rule banning fake reviews and testimonials
- Advertising Standards Authority (UK), CAP Code section 2: Recognition of marketing communications