An AI disclosure label is a visible notice, usually a short phrase, badge or overlay, telling the person looking at an advertisement or post that generative artificial intelligence (AI) created or materially altered what they are seeing. It exists because image, video and voice models now produce material that most viewers cannot tell apart from a photograph or a recording, and because regulators, unions and platforms decided that the burden of telling them should fall on whoever publishes it. Behind it sit advertiser declarations, platform settings and machine-readable marks. As of November 2026, it is mandatory for at least some advertising in the European Union, India, China and New York, and Google applies it automatically in several more markets.

How a label reaches the screen

A label is the last of four steps, and each one is run by a different party.

  1. Creation. A model provider, or the ad platform's own generative tool, produces an image, video, voice track or piece of text. Many now embed an invisible mark at this point: Google's SynthID watermark, or Coalition for Content Provenance and Authenticity (C2PA) metadata that records the asset's digital source type.
  2. Declaration. The advertiser or agency states whether the creative contains AI material. In Google's systems, this is a setting rolled out in July 2026 across Google Ads, Display & Video 360, Campaign Manager 360, Merchant Center and Ads Editor. In the Display & Video 360 (DV360) and Campaign Manager 360 APIs the switch is a field called syntheticContentAttestationStatus, set to IS_SYNTHETIC or NOT_SYNTHETIC, and once set, it cannot be returned to unset.
  3. Detection. The platform reads embedded marks and may run its own classifiers. Assets produced with Google's generative ad tools are flagged automatically, and SynthID and C2PA markup are written into them independently of the setting advertisers control.
  4. Display. The platform renders the label: a caption, a corner overlay, a tag under the account name, or an entry in an ad-information menu. Google added a How this ad was made panel to its ad menus on July 9, 2026, which appears only when an ad contains AI-created or AI-edited assets.

The distinction between steps 1 and 4 matters. A watermark or metadata tag is a machine-readable mark; it is invisible and can be stripped. A disclosure label is a visible statement to a human. Under an IAB Austria guide, neither C2PA metadata nor SynthID amounts on its own to the visible label that a qualifying ad needs, and the same guide finds that the agency, not the client, usually carries the duty under EU law.

Wording varies by platform and by law. Meta uses "AI info". TikTok applies an AI-generated content (AIGC) label. YouTube asks creators to declare "altered or synthetic" content. California's statute, by contrast, prescribes language substantially similar to "this performance features a synthetic performer".

Origins: from political ads to every format

The first ad-specific rules targeted elections. Google announced on September 6, 2023 that election advertisers would have to disclose synthetic content that depicts real people or events, according to Axios, with the policy taking effect in November that year. TikTok launched its creator AIGC label on September 19, 2023.

Labelling then spread to ordinary content and commerce. Google Merchant Center began requiring the IPTC trainedAlgorithmicMedia tag on AI-generated product images in February 2024. On May 9, 2024, TikTok became the first video platform to read Content Credentials and label imported content automatically. Meta's experience showed how hard the wording is: its "Made with AI" label, introduced in May 2024, attached itself to genuine photographs retouched with tools such as Adobe's Generative Fill. After photographers complained, Meta renamed it "AI info" on July 1, 2024. In February 2025 Meta extended AI labelling to ads made or significantly edited with its own generative tools, according to the company's newsroom.

Regulation arrived in 2025 and 2026. China's labelling measures, issued on March 14, 2025 by the Cyberspace Administration of China and three other agencies, took effect on September 1, 2025, requiring both explicit (visible) and implicit (metadata) labels on AI-generated images, audio and video. YouTube made disclosure of realistic synthetic content mandatory on May 21, 2025. India's Ministry of Electronics and Information Technology published draft amendments to its IT Rules in October 2025 that proposed labels covering at least 10% of a visual's surface; the rules notified in February 2026 settled on a requirement for prominent labels, according to Business Today and other Indian press reports.

New York's General Business Law section 396-b, signed on December 11, 2025, took effect on June 9, 2026. It requires a conspicuous disclosure when an advertiser has actual knowledge that a synthetic performer, a generated human resembling no real person, appears in an ad, with civil penalties of $1,000 for a first violation and $5,000 for each one after. Audio-only ads are excluded.

The European Union's AI Act became the largest single driver. Article 50 has applied since August 2, 2026. It obliges providers of generative systems to mark output in a machine-readable form and deployers to disclose deep fakes visibly. Commission guidelines and a Code of Practice published on July 20, 2026 deny advertising the lighter regime for artistic works, and fines reach EUR 15 million or 3% of worldwide turnover. Google signed the code on July 24 and Meta four days later.

The industry's threshold test

Not every use of AI triggers a label, and the hardest question is where the line sits. The IAB's AI Transparency and Disclosure Framework, first released in January 2026, paired consumer-facing labels with C2PA metadata and listed synthetic avatars that simulate human interaction among uses needing disclosure. Version 2, issued on August 18, 2026, set out examples: a photorealistic AI influencer in paid social or display needs a label, an animated mascot does not, and nor do incidental AI-generated figures in a crowd. It also defined two custom C2PA assertions, com.iab.threshold, recording whether AI use crossed the labelling threshold, and com.iab.disclosure, recording whether a label was shown. The framework is voluntary.

Why it matters to marketers

The label carries a measurable cost. An NYU Stern study cited by the IAB found that disclosing generative AI use cut an ad's click-through rate by 31.5%. The IAB's January research found that 39% of Gen Z consumers felt negative about AI ads, against 20% of Millennials. A Raptive survey of 3,000 US adults found trust in content fell by nearly half when readers believed it was AI-made. That gives advertisers a reason to read exemptions narrowly.

Liability is the second issue. Google's setting is an advertiser attestation, and PPC Land has reported that Google shifts AI ad labelling liability entirely to advertisers. A wrong declaration in an irreversible API field cannot be quietly corrected, so agencies need a record of which tools touched each asset.

The label also does not launder a prohibited ad. Google has clarified that AI labels won't save public figure impersonators from Google Ads suspension: an AI-generated likeness of a real public figure used to deceive remains banned whether or not it is disclosed.

Limitations and disputes

Compliance is thin where it is not enforced. A Resemble AI scan of 37,768 Facebook and Instagram ad files from September 2026 flagged 3,379 as AI-generated and found that only 2 mentioned AI in the ad copy. The vendor published no error rates for its own detector.

Labels depend on marks surviving distribution. When The Washington Post uploaded an AI-generated video to eight platforms in October 2025, only YouTube disclosed anything. "No platform that The Post tested kept the Content Credentials data on the video or let users access it," wrote Kevin Schaul. The IAB framework concedes that platforms render advertiser-encoded C2PA mainly for on-platform content, not for third-party creative served through programmatic pipes.

Over-labelling is the opposite failure, as Meta's 2024 rename showed. Placement is contested too: YouTube moved its labels in May 2026 to positions viewers would actually see, an admission that earlier placements were easy to miss.

Legal standards remain undefined. New York does not define "conspicuous" and prescribes no wording; California prescribes wording but applies only when a synthetic performer is used "prominently". In the EU, Article 50(4) speaks of deep fakes, defined as content resembling existing persons, objects, places or events, so whether a wholly invented spokesperson in an ad needs a visible label is answered by the Commission's guidelines rather than the text itself.

What it is not

  • Watermark or provenance metadata. SynthID and C2PA are machine-readable marks that can trigger a label but are not one.
  • AI detection. A classifier that estimates whether content is synthetic informs labelling decisions; it does not disclose anything to the viewer.
  • Paid-for or sponsorship disclosure. "Sponsored" and "Paid for by" labels identify who paid, not how the content was made. A single ad can need both.

Recent developments

The map of mandatory labels keeps widening. PPC Land reported that publishers face Google AI ad labels in California, South Korea and Vietnam, adding to the EU, India and New York overlays. California's SB 1050, signed on September 16, 2026 according to law firm McDermott Will & Schulte, applies from January 1, 2027 with penalties of up to $2,500 per violation. Pennsylvania enacted an Artificial Intelligence Deceptive Advertising Act with disclosure requirements for AI-generated advertising, according to the Consumer Finance Monitor blog of law firm Ballard Spahr, published on November 2, 2026. The next EU milestone is December 2, 2026, when generative AI systems already on the market before August 2 must bring machine-readable marking into conformity.

Timeline

  • September 6, 2023: Google announces mandatory synthetic content disclosure for election ads
  • September 19, 2023: TikTok launches its creator AIGC label
  • February 2024: Google Merchant Center requires the IPTC trainedAlgorithmicMedia tag on AI-generated product images
  • May 2024: Meta introduces the "Made with AI" label
  • May 9, 2024: TikTok begins auto-labelling content carrying C2PA Content Credentials
  • July 1, 2024: Meta renames "Made with AI" to "AI info"
  • February 2025: Meta extends AI labelling to ads made with its generative tools
  • March 14, 2025: China issues its AI-generated content labelling measures
  • May 21, 2025: YouTube's synthetic content disclosure becomes mandatory
  • September 1, 2025: China's labelling measures take effect
  • October 2025: India publishes draft IT Rules amendments on synthetic content
  • December 11, 2025: New York's synthetic performer disclosure law is signed
  • January 2026: IAB releases its AI Transparency and Disclosure Framework
  • February 2026: India notifies its amended IT Rules
  • June 9, 2026: New York General Business Law section 396-b takes effect
  • July 9, 2026: Google adds AI label settings across five ad products and the How this ad was made panel
  • July 20, 2026: European Commission publishes Article 50 guidelines and the Code of Practice
  • August 2, 2026: EU AI Act Article 50 transparency obligations apply
  • August 18, 2026: IAB issues version 2 of its disclosure framework
  • September 16, 2026: California SB 1050 is signed
  • November 2, 2026: Pennsylvania's AI deceptive advertising law is reported as enacted
  • December 2, 2026: EU conformity deadline for generative systems already on the market
  • January 1, 2027: California SB 1050 takes effect

Summary

Who: Ad platforms including Google, Meta, TikTok and YouTube, which design and render the labels; advertisers and agencies, which declare AI use and carry most of the liability; regulators in the EU, China, India, New York, California and Pennsylvania; and industry bodies such as the IAB.

What: A visible notice telling viewers that an ad or post was generated or materially altered by AI, distinct from the invisible watermarks and metadata that often trigger it.

When: Ad-specific rules began with Google's election ad policy in 2023, spread across social platforms in 2024 and 2025, and became law in China (September 2025), New York (June 2026) and the EU (August 2026), with California following on January 1, 2027.

Where: On ad units, posts and ad-information menus across search, social, video and shopping surfaces, with mandatory overlays tied to the location a campaign targets.

Why: Generative models produce images, video and voices viewers cannot reliably distinguish from real ones, and lawmakers want people to know when what they are being sold was never photographed or recorded.