A Performance Max campaign can generate a product image, serve it, and attach an AI disclosure to it without anyone on the marketing team making a decision. The disclosure is not a setting someone forgot to check. In that situation Google applies it on your behalf, and you cannot remove it.
That is the part of Google’s AI ad labelling rollout that has not been written about properly. Most coverage framed it as a transparency feature. For anyone actually running accounts, it is closer to a metadata problem: your production chain now decides what your ads say about themselves.
What Google actually switched on
Google announced the change on 9 July 2026 and rolled the controls out through July across Google Ads, Display & Video 360, Campaign Manager 360, Merchant Center and Ads Editor. Ads across Search, YouTube and Discover now carry a “How this ad was made” section inside the My Ad Center panel, reachable from the three-dot menu or the info icon on the ad. If an asset is designated as AI created or edited, that panel says so.
The setting itself is simple enough. In the Asset library there is an “AI label” status column, and during campaign creation you get a “Review assets” prompt asking you to mark each asset as created or edited with AI, or not. Google’s documentation on AI content label settings spells out the steps.
What that documentation also says, in a note most people skim past, is that the setting is not always yours.
Three ways you get labelled without choosing
- Google’s automation made the asset. When you use its fully automated features to create and serve creatives, assets may be labelled on your behalf.
- Google is legally required to label in that region. Where local law puts the duty on Google, Google acts.
- A signal arrived from somewhere else. Google embeds SynthID watermarks and C2PA provenance markup into images and videos generated inside its own tools, and it reads comparable signals coming from other platforms.
In all three cases, Google’s guidance states plainly that the labels cannot be overwritten.
The third route is the one that catches agencies. An image generated in a third-party tool that writes C2PA metadata arrives carrying its own provenance record. You upload it, you answer “don’t label this asset”, and the label can still appear, because the file is talking about itself.
Where the label is visible, and where it is buried
This distinction decides how much anyone actually notices.
In most of the world the disclosure lives inside My Ad Center, behind the three-dot menu. Realistically, almost nobody opens that. For campaigns targeting the European Union, India and New York, ads with assets designated as AI created or edited also carry a visible overlay on the ad itself.
So the same creative can run quietly in one market and wear a badge in another. If you run pan-European campaigns out of a US account, that difference is not cosmetic. Search Engine Land’s report on the rollout includes the screenshot Google shared, which is worth looking at before you try to describe the panel to a client.
The workflow trap: your own label can get cropped
Google permits you to burn your own AI label into a creative, and confirms this will not violate its policies on text overlays and watermarks. Two practical warnings come attached.
First, keep the label away from corners and edges, because rendering across formats can trim it. Second, and this is the one that bites, Google’s own image enhancements can crop your label. If you want a hand-placed label to survive, you have to opt out of Asset Enhancements in Display and Performance Max campaigns, and Adaptive Layouts in Demand Gen.
That is a genuine trade-off. Those enhancements exist because they usually improve performance. A four-person ecommerce team running Performance Max now has to choose between automated creative optimisation and control over its own disclosure. There is no configuration where you get both.
What to check before your next campaign
Concrete and boring, which is the point:
- Add the “AI label” status column to your asset reporting. You cannot manage a status you cannot see.
- Audit assets you inherited. Stock libraries and freelancer deliveries from the past year may carry provenance metadata nobody logged at the time.
- Write down which markets each campaign reaches, because visible overlays apply only to some of them.
- Decide once, in writing, whether hand-placed labels are worth losing asset enhancements. Then stop relitigating it campaign by campaign.
- Tell clients before they find out from a screenshot. An agency explaining a label after the client spots it is having a much worse conversation.
What can go wrong, and when this barely matters
The failure mode here is not a policy violation. It is a surprise. A brand that has told its audience it does not use AI in creative, then ships a Performance Max campaign built on generated assets, has a credibility problem that no setting fixes. It is the same underlying gap as unapproved AI tools spreading through a marketing team: the org chart says one thing and the file metadata says another.
The opposite risk is over-correction. Google states directly that using its AI label setting does not guarantee compliance with any specific regulation. Treating the toggle as your compliance programme is a mistake, particularly if you serve the EU, where the disclosure duty sits on you rather than on Google.
This may not be worth much of your attention if you run text-only Search ads with no generated imagery, no automated creative features and no third-party AI anywhere in the pipeline. That describes fewer accounts every month, but it still describes some. If it describes yours, note the change and move on.
For everyone else, the shift worth making is this. Stop treating AI disclosure as a communications decision taken at campaign level. It is now a property of your asset pipeline, set upstream by whichever tools touched the file. Audit the pipeline, not the campaign.
Editor’s note: This area changes quickly, so check the latest platform policy before making compliance decisions.