The IAB published version 2 of its AI Transparency and Disclosure Framework on August 18, 2026, setting out which uses of generative AI in advertising creative warrant a consumer-facing label and which do not. The document leans on an NYU Stern finding that disclosing generative AI involvement cut an advertisement's click-through rate by 31.5 percent, a cost the trade body uses to argue against blanket labeling.

The framework arrives sixteen days after transparency obligations under Article 50 of the European Union's AI Act became applicable, and roughly ten weeks after New York's synthetic performer disclosure law entered into force. It replaces guidance the IAB first issued in January 2026, and it is voluntary. According to the IAB, the document carries no enforcement authority and exists to help the industry self-regulate.

That framing matters because the trade body is not writing on a blank page. Three binding regimes already reach advertising creative, and they disagree with each other and, in two specific cases, with the IAB.

What triggers a consumer label

The framework rests on a single test: does the AI involvement create a material risk that consumers will be misled about what they are seeing, hearing, or interacting with. Materiality, in the document's definition, refers to AI use that could mislead a reasonable consumer about what is authentic, factual, or human-created. The threshold is consumer impact, not the volume of AI tools in the production chain.

Three evaluative criteria sit beneath that test. Deception potential asks whether a reasonable consumer could be misled about the nature, origin, or authenticity of what they are viewing. Material impact asks whether AI involvement affects product representation, performance claims, or social proof in ways that influence purchase decisions. Expectation alignment asks whether the content falls outside what consumers would reasonably expect from standard creative production.

Images and video

Disclosure is required for images generated from AI prompts, whether text-to-image or image-to-image, regardless of subsequent human refinement, editing, or compositing. Obviously non-realistic content is excluded. The same logic applies to video generated from text-to-video, image-to-video, or video-to-video prompts.

Disclosure is not required for routine post-production analogous to established editorial practice: color correction, lighting adjustment, background cleanup, dust and blemish removal, contrast optimization. Nor for clearly stylized or fantastical imagery, or for technical improvements such as upscaling, de-noising, and format conversion that do not alter content meaning.

The document works through examples. An e-commerce retailer generating furniture images from prompts without photographing either the product or the room requires disclosure. A supplement company placing a real product photograph against AI-generated kitchen and gym backgrounds does not. Nor does an automotive spot using AI to remove safety rigging from a stunt sequence, on the grounds that this is standard visual effects practice.

Audio and voice

Two audio cases trigger disclosure. The first covers synthetic voices of deceased persons creating statements they never made, and applies even where an estate has granted authorization. The second covers synthetic voices of living persons depicting events, actions, or circumstances that never occurred, in a manner a consumer could mistake for authentic documentation.

The carve-out is broad. Authorized voice clones used for scripted commercial content, generic synthetic voiceovers where speaker identity is immaterial, AI-generated background music, noise reduction, equalization, and pitch correction all fall below the threshold.

One example marks the boundary precisely. An automotive company using an authorized AI voice clone of a racing champion, narrating in the first person a dramatized recreation of a race he genuinely won as though recorded at the track that day, requires disclosure. The authorization is real and the underlying event happened. The recreation is what triggers the label.

Synthetic influencers and digital twins

Disclosure is required for digital twins of deceased individuals used in any capacity, for digital twins of living individuals shown in fabricated events or locations, and for AI chatbots or conversational agents that engage consumers in ways simulating human interaction. Conversational agents carry a distinct label, AI-powered, rather than the standard AI-generated wording.

A photorealistic AI influencer run as paid social and display creative requires disclosure. An animated gecko mascot does not. Nor do incidental AI-generated figures in crowds or background scenes where individual authenticity is not material to the message.

Text and copy

For text and copy, disclosure is generally not required. Headlines, slogans, product descriptions, email subject lines, blog posts, and translations sit outside the framework's labeling scope. Advertisers remain responsible for accuracy and substantiation regardless of authorship method, and the document is explicit that AI-generated false claims or unverified data remain prohibited under existing consumer protection law, with disclosure providing no cure.

The sparkle icon and the two-layer model

For content that meets the threshold, the framework accepts two consumer-facing methods: a standardized visual indicator or a plain text label reading "AI-generated." Either fully satisfies the requirement. For the icon, the IAB adopted the Unicode sparkle at codepoint U+2728, rendered as a monochrome glyph. The reasoning is practical rather than aesthetic. The character requires no licensing, is available to every advertiser, agency, platform, and publisher, and scales across formats.

Placement rules attach. Visual labels near an image where feasible; video labels on the first frame, persisting throughout; audio disclosures spoken before or immediately after the AI-generated segment, repeated at least once if the advertisement exceeds 60 seconds. Text labels must meet the Web Content Accessibility Guidelines AA contrast ratio of 4.5:1, include alt text for screen readers, and appear in captions.

Underneath the visible layer sits a machine layer built on Coalition for Content Provenance and Authenticity credentials. The IAB is defining two custom assertions for the standard: com.iab.threshold, recording whether AI involvement met the disclosure threshold, and com.iab.disclosure, recording whether a consumer-facing label was applied. Metadata also captures involvement type and tool identification. Timestamps, tool versions, and signer entity can be captured automatically. The disclosure classification itself remains a human judgment.

The document acknowledges the weakness. Metadata is stripped when files are re-saved, compressed, or screenshotted. Embedded watermarking, Google's SynthID among them, is treated as a reinforcing complement rather than a substitute.

Accountability sits with the advertiser

Ultimate responsibility rests with the organization that creates the advertising content, and the framework states plainly that this accountability cannot be delegated. Where agencies create and control content for clients, agency and client may share it.

Platforms serve as secondary enforcers. Where an advertiser relies on a platform to render a label from C2PA metadata, the advertiser remains accountable for the outcome; if the platform fails to render, the obligation has not been met. The document notes a live gap here. Platform rendering of advertiser-encoded C2PA into consumer-facing labels currently works mainly for content generated on-platform, not for third-party creative arriving with metadata attached.

That allocation matches what has already happened commercially. Google's July 2026 documentation placed the disclosure duty on advertisers rather than on AdSense publishers, giving advertisers an AI label setting across five products and a consumer-facing How this ad was made panel inside My Ad Center. Display and Video 360 and Campaign Manager 360 gained an attestation field that cannot be unset once written.

Where the framework departs from EU law

The IAB identifies two commercial-endorsement cases where its thresholds sit below Article 50(4) of the AI Act: authorized voice clones of a living person, and authorized digital twins of a living person not placed in fabricated events. The reasoning is that traditional endorsement framing removes materiality risk, because a consumer watching a celebrity endorsement understands they are seeing paid commercial messaging.

Article 50(4) contains no such carve-out. AI-generated audio or video resembling an existing person and appearing authentic falls within its disclosure scope regardless of consumer expectation. The framework instructs advertisers operating in EU markets to yield to the stricter requirement. South Korea, Vietnam and India likewise decline to recognize the endorsement exemption.

Non-compliance with Article 50 carries administrative fines reaching 15 million euros or 3 percent of worldwide annual turnover, whichever is higher, and the European Commission's implementation guidelines exclude persuasive commercial content from the lighter disclosure regime available to artistic or satirical work.

A fragmented regulatory map

The framework's regulatory section states policy as of August 3, 2026, and the picture it describes is uneven. Over 1,000 AI-related bills were introduced across United States states in 2025 alone. New York's synthetic performer law took effect June 9, 2026, carrying civil penalties of $1,000 for a first violation and $5,000 for each subsequent one. California's AI Transparency Act took effect August 2, 2026, but binds covered providers with over one million monthly users rather than advertisers, and the visible label it enables is optional with no direct penalty on brands for omitting it.

At federal level, a December 2025 Executive Order directed the government toward a uniform national AI policy framework and challenges to inconsistent state laws. The Department of Justice's AI Litigation Task Force began challenging state AI laws in court in April 2026. No preemption statute has been enacted, and state disclosure laws remain enforceable.

Across Asia, China's labeling measures took effect September 1, 2025. South Korea's AI Basic Act took effect January 22, 2026 with a one-year enforcement grace period, and a pending Network Act amendment would require labeling of AI-generated photos and videos regardless of deception risk, with punitive damages up to five times actual losses for malicious distribution of false content. Vietnam's AI law came into force March 2026, with visible labels required under an implementing decree effective May 2026. India's amended IT Rules mandate labeling of synthetically generated information. Japan, Singapore, and Australia favor voluntary approaches.

Two dates that do not agree

The two documents released by the IAB give conflicting accounts of the EU Code of Practice on Transparency of AI-Generated Content. The framework states the Code was finalized in June 2026 and "provides official EU icons and design and placement rules." The accompanying announcement states the Code was finalized in early June 2026 and "includes illustrative examples of a possible common EU icon that has not yet been finalized."

Those are different claims about the same instrument. The framework separately notes that the Commission adopted final Guidelines on Article 50 on July 20, 2026. PPC Land reported that the Commission published its Article 50 guidelines and the finalised Code together on July 20, 2026, and separately released a free set of three labelling icons with a last-update date of 10 June 2026. The framework also places the January 2026 predecessor without a day; PPC Land covered that release on January 16, 2026. Advertisers reconciling the two IAB documents against the Commission's own timeline will find the icon question unsettled in the source material itself.

Roughly 190 companies and organizations had signed the Code by the end of July 2026. Google signed on 24 July; Metafollowed days later, reversing a 2025 refusal.

The evidence the framework rests on

Deployment numbers are not in dispute. The IAB and Sonata Insights study published in January 2026 found 83 percent of advertising executives reporting AI deployment in the creative process, up from 60 percent in the 2024 edition. An Advertising Perceptions omnibus survey from March 2026 put 91 percent of advertisers as currently using AI for creative development or planning to.

The sentiment numbers are where the argument sits. Eighty-two percent of advertisers believe consumers feel positive about AI-generated advertisements, against 45 percent of Gen Z and Millennial consumers who actually do, a 37-point gap that has widened since 2024. Cost efficiency is now the top-cited benefit among advertisers, up from fifth place two years earlier. Pew Research found 76 percent of United States adults consider it extremely or very important to distinguish AI-made content from human-made content. Among Gen Z, 34 percent regard brands using AI in advertising as "creative" while 30 percent consider them "inauthentic."

Against that, 73 percent of Gen Z and Millennial consumers said knowing an advertisement was created with AI would either increase or have no effect on purchase likelihood. Seventy-two percent of advertisers surveyed by Advertising Perceptions want industry standards for disclosure, but only 17 percent believe AI needs disclosing for every instance of use.

The NYU Stern figure is the counterweight the framework returns to. Telling consumers an advertisement was made with generative AI cut click-through by 31.5 percent. The document's position is that this cost is warranted where AI use could mislead, and unwarranted where it could not, which is the entire basis for rejecting blanket labeling. It also raises the implied-truth effect: audiences perceive unlabeled content as more credible simply because other content carries a label.

Separate research points the same direction from a different angle. A Raptive study of 3,000 United States adults measured a near-halving of reader trust where content was suspected of being AI-generated, with a 14 percent decline in purchase consideration for adjacent advertising.

What the working group said

Caroline Giegerich, VP, AI at the IAB, led the working group. "Trust is everything between a brand and its customers, and being honest about AI is part of earning it," she said. "That said, not every use of AI needs a label - labeling everything teaches consumers to ignore labels and could negatively impact advertisers. This is why we take a meticulously nuanced position in this framework."

Giegerich framed the drafting problem as bounded on two sides. "The working group's central challenge was avoiding two failure points," she said. "Under-disclosure leaves consumers at risk of being misled. Over-disclosure could risk negatively impacting advertisers."

Graham Wilkinson, EVP, Chief Innovation Officer and Global Head of AI at Acxiom, located the commercial case in the pace of regulation. "Regulators in New York, California, South Korea and the EU are no longer asking whether AI disclosure matters, they're setting enforceable rules for it," he said. "An industry framework that keeps pace with that shift, rather than reacting to it state by state, is what lets us keep innovating without leaving consumers or regulators guessing."

Simon Poulton, EVP, Innovation and Growth at Tinuiti, described the agency position. "Agencies sit at the center of this conversation, translating this framework into what actually runs in-market for our clients," he said. "Having clear, consistent guidance on disclosure means less guesswork and less risk for the brands we represent, and that benefits the entire industry."

Michael Lampert, Senior Director, Global Gen AI Adoption and Consumer Data Strategy at Mondelez International, put the gap at the center. "There is data that shows consumers are open to trusting AI in advertising, but only when brands are upfront about it," he said. "As adoption accelerates across the industry, that gap between AI use and consumer trust is the one that this framework is built to help close."

David Cohen, CEO of the IAB, said: "Trust is a foundational element which is critical to the growth of AI across the ecosystem. Putting the right transparency and disclosure standards in place benefits everybody: consumers, advertisers, agencies, publishers and platforms."

Implementation runs 24 months

The roadmap has three phases and the IAB is explicit that it is not a compliance schedule. Organizations facing binding requirements in the EU, New York, or APAC markets must meet the legal deadline regardless of position in the roadmap.

Phase one covers zero to six months: designate an AI Disclosure Lead within 60 days, train creative, production, media and communications teams, add a four-question pre-launch checklist to existing workflow tools, and fold disclosure review into legal review rather than creating a separate approval layer. Phase two, months six to twelve, runs controlled pilots measuring label noticeability, click-through and conversion effects, and the time disclosure decisions consume. Phase three, months twelve to twenty-four, scales the practice and gives creative management systems and ad servers the capability to flag assets and append disclosure language automatically.

Dynamic creative and real-time personalization get separate treatment. Where components exist before the campaign, each is evaluated and any triggering component pulls the disclosure onto the assembled creative. Where components are generated at serve time, the advertiser defines disclosure rules at the system level and the generative system applies them per output. The framework cites a Cadbury campaign that produced 130,000 unique AI videos for local Indian businesses as the scale problem this addresses.

Why this matters for marketing

The practical value of the framework is not the sparkle icon. It is the list of things that do not need a label. Advertisers running AI-assisted production at volume have been operating without a defensible line between assistive editing and synthetic creation, and three regulators have drawn that line differently. A shared industry baseline gives agencies and brands a single internal standard to apply, then adjust upward where local law is stricter.

The exposure is uneven. A brand running programmatic display across the EU, CTV in the United States, and paid social in Asia now faces platform labels from Meta, TikTok and Google that each define their trigger differently, alongside Article 50, New York's statute, and India's IT Rules. The IAB framework does not resolve that fragmentation. It offers a position to negotiate from.

The unresolved piece is the endorsement exemption. An advertiser using an authorized celebrity voice clone can skip the label under IAB guidance and must apply it under EU law, which means the exemption is only usable in markets where nothing stricter applies. Whether that carve-out survives contact with regulators, or quietly becomes a distinction advertisers stop drawing, is the question the next version of this document will have to answer.

Timeline

  • July 15, 2025 - Raptive publishes research finding suspected AI content cuts reader trust by nearly half and purchase consideration by 14 percent
  • September 1, 2025 - China's Labeling Measures for AI-Generated Synthetic Content take effect
  • December 2025 - A United States Executive Order directs the federal government toward a uniform national AI policy framework
  • December 11, 2025 - New York's synthetic performer disclosure law is signed
  • January 16, 2026 - The IAB publishes version 1 of its AI Transparency and Disclosure Framework alongside Sonata Insights research documenting a 37-point perception gap
  • January 22, 2026 - South Korea's AI Basic Act takes effect with a one-year enforcement grace period
  • February 20, 2026 - India's amended IT Rules take effect, including synthetic content labelling duties
  • March 2026 - Advertising Perceptions finds 91 percent of advertisers using or planning to use AI for creative development
  • April 2026 - The Department of Justice's AI Litigation Task Force begins challenging state AI laws in court
  • June 2026 - The EU AI Omnibus package gives systems on the market before August 2, 2026 until December 2, 2026 to meet marking obligations
  • June 9, 2026 - New York's synthetic performer disclosure requirement enters into force
  • June 10, 2026 - The European Commission publishes free AI labelling icons
  • July 9, 2026 - Google introduces an AI label setting across five advertising products
  • July 13, 2026 - Google documentation confirms the disclosure duty sits with advertisers rather than AdSense publishers
  • July 20, 2026 - The Commission adopts final Guidelines on Article 50 and publishes the finalised Code of Practice
  • July 24, 2026 - Google signs the Code of Practice on Transparency of AI-Generated Content
  • End of July 2026 - Roughly 190 companies and organizations have signed the Code
  • August 2, 2026 - Article 50 transparency obligations and California's AI Transparency Act both take effect
  • August 3, 2026 - The state of policy reflected in the framework's regulatory section
  • August 18, 2026 - The IAB publishes version 2 of the AI Transparency and Disclosure Framework

Summary

Who: The Interactive Advertising Bureau, through a cross-industry working group led by Caroline Giegerich, VP, AI at the IAB, with members from Digitas, Acxiom, Conde Nast, Mondelez International, Tinuiti and Warner Brothers Discovery. The guidance addresses advertisers, agencies, publishers, platforms and technology partners.

What: Version 2 of the AI Transparency and Disclosure Framework, a voluntary risk-based model specifying which uses of generative AI in advertising creative require a consumer-facing label. Disclosure is triggered for synthetic images and video generated from prompts, synthetic voices of deceased persons, synthetic voices of living persons in fabricated scenarios, synthetic avatars, digital twins of deceased individuals, digital twins of living individuals in fabricated events, and conversational agents in advertisements. Routine post-production, stylized characters, authorized commercial voice clones, generic voiceovers, background music, and text and copy fall outside the labeling requirement. The document accepts either the Unicode sparkle at U+2728 or a plain text label, and pairs the visible layer with C2PA metadata carrying two new IAB assertions.

When: Published August 18, 2026, sixteen days after Article 50 of the EU AI Act became applicable on August 2, 2026, and roughly seven months after the January 2026 first version. The implementation roadmap runs 24 months, with an AI Disclosure Lead designated within 60 days.

Where: Issued from New York and applicable to advertising and marketing communications globally, with explicit deference to stricter binding law in the European Union, New York, California, South Korea, Vietnam and India.

Why: Inconsistent disclosure practice across the industry is producing both consumer confusion and label fatigue while three separate regulatory regimes impose divergent requirements on the same creative. An NYU Stern finding that disclosure cut click-through by 31.5 percent supplies the framework's case against blanket labeling, set against a 37-point gap between what advertisers assume consumers feel about AI-generated advertising and what consumers report.