Google will start telling users when an advertisement was created or edited with artificial intelligence.
A new panel called "How this ad was made" is rolling out globally across Search, YouTube and Google Discover, reached through the three-dot menu or info icon already attached to ads.
It marks the first time Google has extended AI disclosure to commercial advertising, having previously required it only for political ads since 2023.
The reasons behind the move say more than the feature itself.
The trust problem
The immediate driver is a straightforward risk to trust.
Generative AI has made it cheap to produce slick, photorealistic product imagery, and shoppers can easily mistake a synthetic scene for a genuine photograph.
If enough advertising turns out to be fabricated without disclosure, users start to doubt everything they see, which erodes the value of the platform Google sells to advertisers.
Labelling is a way to protect that trust before it frays.
Google's own tools are the cause
There is a more pointed reason, and it sits inside Google itself.
The company now offers advertisers generative AI tools for making text, images and video, and embeds its Gemini models across products such as Performance Max and AI Max.
Google is therefore a major supplier of exactly the synthetic content the label is meant to flag.
Ads built with its own AI tools will be disclosed automatically, with no action required from the advertiser.
That automatic labelling is possible precisely because Google controls those tools and knows when they were used.
The regulatory clock
The timing is not accidental.
The European Union's AI Act imposes transparency obligations on realistic synthetic media, and its Article 50 rules become enforceable on 2 August, with penalties reaching 15 million euros or 3% of global annual turnover.
The United States Federal Trade Commission has signalled parallel scrutiny of AI in advertising, and New York has passed its own law on digital replicas of real people.
By building one disclosure system now, Google front-runs the regulation, hands advertisers a ready-made compliance tool, and avoids maintaining separate ad ecosystems for each jurisdiction.
On-ad labels will appear directly in markets that require them, including the European Union, India and New York.
There is a competitive angle too.
Meta has faced criticism over AI content across Facebook and Instagram, and Amazon over AI product images, so Google gains by positioning itself as the responsible operator while trust is scarce.
The gap in the plan
The weakness lies in everything Google does not control.
When an advertiser uses a third-party tool such as a rival image generator, disclosure depends on the advertiser ticking a box, and Google has said it will not verify whether that box is ticked honestly.
That leaves the label resting on an honour system, and the incentive runs the wrong way.
An advertiser hoping a synthetic image passes for a real photograph has little reason to volunteer otherwise.
Rivals Meta and TikTok already read embedded provenance data to detect and label third-party AI content automatically, so Google's reliance on self-disclosure looks weaker by comparison.
Google is not even consistent across its own products, since on YouTube it auto-labels AI-generated video whether or not the creator discloses it.
The resistance already forming
The mandatory version of these rules is meeting pushback.
Retailers are lobbying to exempt AI-assisted commercial ads from the EU transparency requirements, arguing routine creative should not be treated as deceptive synthetic media.
Google has been careful to note that its label does not guarantee compliance with any specific law, pushing legal responsibility back onto advertisers.
What it signals
Taken together, the move shows AI disclosure shifting from a narrow political-ad rule to a standard feature of commercial advertising.
The label is real and useful, but its coverage depends on which tools an advertiser used and whether they chose to be honest.
For ads made with Google's own systems it is automatic and reliable, and for everything else it is a switch that the honest will flip and the rest will quietly ignore.
That gap is both the feature's limitation and, tellingly, the reason it was needed.
