Transparency Disclosure Norms for AI-Generated Brand Content

AI-generated brand content now requires a jurisdiction-aware disclosure framework for deciding when and how to use a “Made with AI” label. Four overlapping legal regimes govern different parts of the process, from consumer impressions to embedded markings. Compliance depends on identifying the content, its audience and the brand’s role before choosing how to disclose.

Why four legal regimes now apply simultaneously to AI brand content

The 2026 convergence brings together federal deception rules, dedicated state requirements and European transparency duties. The FTC Endorsement Guides and Section 5 of the FTC Act already apply to misleading AI uses without a standalone federal AI disclosure statute. New York General Business Law Section 396-b, signed December 11, 2025, became effective June 9, 2026. California SB 942 / AB 853 and EU AI Act Article 50 have an August 2, 2026 operative date.

The federal framework addresses deception and endorsements, New York addresses synthetic performers in advertising, California addresses covered AI infrastructure and disclosure mechanisms, and the EU divides responsibilities between providers and deployers. Their boundaries do not align neatly. The EU AI Act Code of Practice on Transparency of AI-Generated Content, published June 10, 2026, provides a compliance roadmap for Article 50. Following it is not conclusive evidence of compliance.

What each of the four regimes actually requires, and where their scopes differ

The FTC framework starts with the impression conveyed to a reasonable consumer. Its Endorsement Guides, updated July 2023, and Section 5 apply when AI involvement misleads people about a real person’s opinion or experience. For sponsored AI content where both issues arise, the disclosure structure has two parts, namely the material connection and the AI involvement. Neither substitutes for the other. Per-violation exposure can multiply across non-compliant campaign posts.

New York’s law has a narrower trigger. It applies to a person who produces or creates an advertisement with actual knowledge that it uses a “synthetic performer.” That term covers a digitally created asset made using generative AI or a software algorithm to give the impression of a human performer who is not recognizable as an identifiable natural person. Conspicuous disclosure is required for advertising targeting New York consumers, regardless of the advertiser’s location. Audio-only advertisements and AI used solely to translate a human performer’s language are excluded.

SB 942 applies to large generative AI providers, while AB 853 also covers large online platforms, generative AI hosting platforms that let California residents download source code or model weights, and the capture device manufacturers named in the law. The framework includes free detection tools, visible manifest disclosures and embedded latent disclosures. Visible and invisible disclosure are separate requirements. California has a law concerning deceased personalities and another concerning labor contracts and digital replica enforceability.

Article 50 gives providers the duty of machine-readable marking and deployers the duty of end-user disclosure, including brands and agencies that publish deepfakes or certain AI-generated public-interest content. Its broader scope also includes direct interaction with individuals, emotion recognition and biometric categorization. European audience exposure can bring a company within scope regardless of its home country. Role allocation therefore precedes disclosure design. A workflow may involve a foundation model provider, model host, platform provider, system integrator and downstream deployer, each occupying a different position in the compliance chain.

Where the four regimes overlap and where they diverge — the compliance collision zones

A sponsored synthetic-performer advertisement targeting New York consumers sits at the intersection of federal and state requirements. The FTC analysis addresses sponsorship and misleading impressions about AI involvement, while New York requires conspicuous disclosure of the synthetic performer. The applicable federal clear-and-conspicuous standard must also be met. One set of disclosures may meet both sets of rules, but an unclear AI hashtag does not show that it does.

European distribution adds a different distinction. For AI-generated images or video covered by the rules, provider marking and any required deployer disclosure do different jobs, so machine-readable marking is still needed when a notice is visible to consumers. California also requires manifest and latent disclosure when its rules cover the provider or platform. Geography alone does not settle every obligation. A brand that builds an AI tool and publishes its output may also occupy both provider and deployer roles, bringing interlocking duties into the same organization.

The gaps are equally important to the map. New York excludes audio-only advertising and language translation alone, but the FTC can still reach a misleading consumer impression in either case. Conversely, California’s embedded watermark requirement does not answer whether an FTC disclosure is clear and conspicuous. The mechanics are cumulative. Questions about particular content, covered entities or supply-chain roles remain fact-dependent and should be resolved with counsel rather than inferred from another jurisdiction’s rule.

The platform enforcement layer brands cannot outsource their compliance to

Platforms add an enforcement layer beyond the four legal regimes. Meta has used mandatory “Imagined with AI” labeling across Facebook, Instagram and Threads since early 2024, combining self-disclosure with algorithmic detection. Its penalties for undisclosed AI usage include reduced visibility, downranking or removal. YouTube requires “Altered or Synthetic Content” disclosure for AI-generated or manipulated media. Campaigns must satisfy legal requirements independently of these platform mechanisms.

Juan Wu, Zhe (James) Zhang and a coauthor’s When Is Self-Disclosure Optimal? Incentives and Governance of AI-Generated Content describes X, Reddit and Bluesky as having no mandatory disclosure regimes. Their analysis also shows that detection accuracy is often below 80% and varies by language, format and adversarial paraphrasing. Detection is therefore an unreliable sole backstop. A platform’s failure to identify synthetic content does not remove the brand’s disclosure obligation, while the absence of a mandatory platform rule does not create a legal exemption.

For brand operations, the implication is channel-specific execution within a common compliance process. Platform-native labels should supplement the disclosures required by the applicable legal regimes. Where self-disclosure operates with detection and penalties, proactive disclosure avoids depending on whether a detector catches an omission.

What the FTC's "clear and conspicuous" standard actually demands in practice

The FTC standard requires disclosure where consumers come across the content. A disclosure hashtag buried among many others fails that test. Back-end metadata or terms of service alone likewise do not supply consumer-visible notice at the point of encounter. The disclosure must address the impression the content creates. That makes placement part of compliance, rather than a finishing decision left to creative preference.

For execution, teams should assess whether a viewer would reasonably believe an AI person is an actual human, a testimonial came from an actual customer, or a realistic scene actually happened. Where AI creates a misleading impression of that kind, the disclosure must address it clearly. A label cannot legalize fake reviews or results. Calling content AI-generated also cannot make a made-up personal experience real.

Sponsored content requires a separate assessment of the material connection. When both sponsorship and AI involvement must be disclosed, teams should make each clear and avoid combining them in one unclear hashtag. An AI label does not explain payment. A sponsorship label does not tell people that a speaker who looks human is synthetic. Both must be clear and conspicuous to the consumer encountering the content, rather than discoverable only after searching elsewhere.

How the EU Code of Practice operationalizes Article 50 for deployers

DLA Piper illustrates the Code’s division of responsibility through hallmarking. The provider acts as the “assay office,” responsible for machine-readable marking and detection capability, while the deployer acts as the “shopfront,” responsible for user-facing disclosure. A mark within the content’s technical structure does not perform the same function as a notice shown to the person viewing it.

The Code’s Section 2 translates deployer duties into a layered approach to labels, icons and disclaimers, including their design and placement. For covered deepfakes and AI-generated text on public-interest matters, disclosure that users can easily see is expected, and metadata alone is not enough. Most publishing brands confront this shopfront responsibility. Brands that also build or host AI tools must separately assess whether provider obligations attach, rather than treating publication as their only relevant activity.

The Code nevertheless guides implementation without conclusively establishing Article 50 compliance. The European Commission published its finalized scope guidelines on July 20, 2026, after the Code’s June 10 publication, when those guidelines remained in draft form. The two instruments answer different questions. The guidelines define applicability, while the Code provides a route for implementing the transparency duties that apply. For deployers, the practical task is to document responsibility across the supply chain before publication, rather than assume a tool supplier has fulfilled every obligation.

Why the trust cost of non-disclosure now exceeds the friction cost of disclosure

The commercial case for disclosure rests on the difference between openly using AI and having concealed use exposed. Wu and colleagues’ model distinguishes voluntary disclosure, detected non-disclosure and strategic concealment, with different engagement outcomes for otherwise identical AI-generated content. It incorporates trust penalties after detected non-disclosure. Within that framework, voluntary disclosure performs better on trust than exposure of an omission, giving transparency a commercial rationale beyond avoiding legal sanctions.

The model offers a theoretical account of brand behavior. It finds disclosure optimal only when both the value of AI-generated content and its cost-saving advantage are at intermediate levels. As AI gets better, the model predicts that enforcement will move from strict deterrence to partial screening and then deregulation. Brands must assess future legal disclosure requirements independently of that modeled trajectory.

The audience case also warrants restraint. Directional consumer evidence supports treating transparency as a potential differentiator for brands with significant Gen Z and Millennial audiences, but it does not establish a fixed trust penalty for every campaign. Disclosure has production friction, while detected concealment can introduce a separate credibility cost. The practical conclusion is to treat transparency as part of the brand’s quality proposition.

A jurisdiction-aware decision framework for when, where, and how to disclose

For implementation, brands should ask four questions in sequence, namely what the content contains, whom it reaches, which regimes apply and what the most demanding applicable standard requires. Content classification comes first. A synthetic human performer calls for a New York assessment when the advertisement targets that state, subject to its audio-only and translation exclusions. AI-generated images, video, audio and public-interest text each need to be checked under the EU provisions that apply. Sponsored content also requires assessment of both the material connection and any misleading impression created by AI involvement.

Audience geography then narrows the applicable rules. The FTC supplies the federal baseline for U.S. consumer-facing deception and endorsements, while New York follows advertising directed at its consumers regardless of the advertiser’s address. California requires a covered-entity analysis as well as geographic attention, including the stated provider threshold of over 1 million monthly California users. European distribution requires an Article 50 scope assessment. Neither an audience location nor the presence of AI alone answers every applicability question.

The disclosure mechanic should follow those determinations. Teams should use clear-and-conspicuous consumer-facing notices where the FTC requires them, conspicuous notices in proximity to covered New York advertising, and prominent deployer disclosures where the EU requirements apply. Machine-readable markings and California latent disclosures need separate treatment from visible labels. Sponsored AI content requires distinct communication of sponsorship and AI involvement where both duties arise. Available Meta and YouTube labels should be used as supplemental mechanisms.

Role identification must be complete before publication. Brands with EU exposure should document whether they build or host the AI tool, publish its output, or do both, and allocate responsibilities across the supply chain. Those decisions belong in creative briefs and production checklists, whether the brand or an agency does the work. This framework is a decision aid. Brands with complex multi-jurisdiction exposure should validate their approach with qualified legal counsel.

Transparent, well-structured public content can provide cleaner signals for AI systems interpreting and representing a brand, while inconsistent disclosure can weaken those signals. Evident’s perception intelligence scoring gives brands a practical way to assess how AI systems see them. It is not a legal compliance determination. Brands need a systematic scoring approach across algorithmic, AI and human evaluation to understand whether their disclosure practices are working.