AI Ad Disclosure Labels: What Brands Must Label In 2026

What Google Changed

In early November 2026 Google updated its help pages to add California, South Korea and Vietnam to the list of locations where ads with AI-created or AI-edited assets show a disclosure on the ad itself. The European Union, India and New York were already on the list. Search Engine Roundtable spotted the change in both the Google Ads and Merchant Center documentation.

The setting behind it arrived in July. Google introduced an AI label field in Google Ads, Display and Video 360, Campaign Manager 360, Merchant Center and Ads Editor. A labelled asset shows a note under "How this ad was made" in My Ad Center wherever the ad runs, and a visible overlay on the ad in the listed locations.

Two Layers Of Disclosure

It helps to separate the two things Google does with a labelled asset.

• Everywhere: the ad carries a note in the "How this ad was made" section of My Ad Center, which a viewer reaches from the three-dot menu on ads in Search, YouTube and Discover. Most people will never open it.
• In the listed locations: the ad also shows a visible overlay on the creative itself. This is the one your audience will see, and the one your design has to leave room for.

Publishers are not part of this. A Google notice to AdSense publishers in July said its sell-side products do not offer AI labelling features, so the label is set by the advertiser in the buying platform, or not at all.

How The Rules Arrived

• August 1, 2024. The EU AI Act enters into force.
• May 21, 2025. YouTube's mandatory AI content disclosure for creators takes effect.
• June 24, 2026. Google Ads API version 24.2 adds structures for labelling synthetic content.
• July 9, 2026. Google publishes the AI label setting across Google Ads, Display and Video 360, Campaign Manager 360, Merchant Center and Ads Editor.
• August 2, 2026. The AI Act's Article 50 transparency obligations begin to apply. A provisional agreement in May delayed the deadlines for high-risk systems but left this date unchanged.
• November 2026. Google adds California, South Korea and Vietnam to the locations with an on-ad label.

Who Is Responsible

The advertiser. PPC Land's reading of the documentation is blunt: Google says neither its automatic label nor a custom label guarantees compliance with any specific regulation, and advertisers are told to take their own legal advice.

Three details matter in practice:

• Labels do not travel between products. Each Google product has its own field. An asset labelled in Google Ads is not labelled in Display and Video 360.
• Google can label for you. It may apply a label where the law requires it, where it receives signals of AI involvement, or where you used Google's own generative tools. Those labels cannot be removed.
• Your own on-image label can be cropped. Google asks for a custom label to be kept at least about 5.5% in from every edge, because rendering can trim the border, and automatic image enhancements can remove it without warning.

The legal backdrop is the EU AI Act. Its Article 50 transparency obligations apply from August 2, 2026, with penalties of up to 15 million euros or 3% of global annual turnover. This article is a production guide, not legal advice, so confirm your own position with counsel.

A Watermark Is Not A Label

Images and video made with Google's own generative tools carry SynthID, an invisible watermark from Google DeepMind embedded in the pixels, along with C2PA provenance metadata. It is built to survive cropping, filtering and compression. It is not a disclosure. A viewer cannot see it, and PPC Land's report is clear that a watermark is not the public-facing notice that transparency laws ask for.

The same goes for the European Commission's free AI labelling icons. They are optional, and using one does not by itself establish compliance with Article 50. Treat watermarks and icons as supporting evidence, set the platform label on every asset, and have counsel confirm what else your markets require.

What Counts As AI-Made

This is where creative teams get stuck. A fully generated scene is obvious. Most real work is mixed.

We produce commercials that combine a real product shoot with AI and 3D environments, such as the Toyota (IMC) Land Cruiser 250 film, where a real car filmed in a studio was placed into locations that were never visited. We also make product and launch films built largely in AI and 3D. Our working rule is simple: if AI created or materially changed what the viewer sees or hears, the asset is labelled. Colour grading, clean-up and resizing are not.

When in doubt, label. A disclosure costs nothing. A missing one in the wrong market is the expensive mistake.

Examples From A Mixed Production

These are the calls we make under that rule. They are our production practice, not a legal standard.

• Real product, generated location. Labelled as AI-edited. The viewer is seeing a place the product was never in.
• Fully generated shot or image. Labelled as AI-created.
• Synthetic or cloned voiceover. Labelled. The rule covers what the viewer hears as well as what they see.
• A generated person or crowd. Labelled, and checked against likeness and talent rules before anything else.
• Clean-up, colour, stabilising and resizing. Not labelled. Nothing the viewer sees has been invented.
• AI used only in planning. Storyboards, scripts and mood frames that never appear in the final asset do not make the asset AI-made.

A Labelling Workflow For Mixed Productions

• Keep a production log. For every deliverable, record which shots, voices and images were AI-created or AI-edited, and with which tool. It takes minutes during production and hours afterwards.
• Put the label in the delivery sheet. Each exported asset gets a column: AI-created, AI-edited or neither. The media team should never have to guess.
• Label in every platform separately. Google Ads, Display and Video 360, Merchant Center, YouTube, Meta and TikTok each have their own control.
• Turn off automatic enhancements on labelled images if your own on-image disclosure must stay visible.
• Agree the market list with legal. The list of locations is growing. Review it each quarter.
• Write it into the brief. Clients should know before production which assets will carry a label and where.

Does A Label Hurt Performance?

Some advertisers worry it will. There is not enough public data yet to say, and it will vary by category. Research from the IAB in January 2026 found a gap between how advertisers and consumers feel about AI in advertising, which is a reason to test in your own category instead of assuming. Our view is that the label matters far less than the work. Viewers respond to whether a film looks finished and whether the product looks real. In the Toyota work the car is real and the craft is visible, and that is what people react to.

The bigger risk is the opposite one: hiding AI use and being called out for it. For how we approach this kind of production, see how AI and 3D are changing brand visuals. For where those ads now appear, see ads inside Google AI Mode.

Sources

• Search Engine Roundtable: Google Ads expands AI content disclosure label to more regions, November 5, 2026
• PPC Land: Google shifts AI ad labeling liability entirely to advertisers, July 14, 2026