Providers mark, deployers label
Article 50 divides its obligations by role. Providers, meaning the companies that build generative systems, carry the machine-readable half: digitally signed metadata plus an imperceptible watermark for audio, images, and video, watermarking for free-form text longer than 200 tokens, and a free detection tool made available to deployers, end users, authorities, researchers, and media.
Deployers carry the visible half. The guidance describes the category as covering agencies, brands, and publishers, and the requirement is a label a person can see on deepfakes and on AI-generated published text, using the EU's icons or equivalent wording reading "AI GENERATED" or "AI MODIFIED". Placement is specified rather than left open: the label appears at first exposure and stays visible before any overlay covers it, at the start of a video and again after ad breaks, and above or near a headline or in a colophon for text.
Enforcement is sized to make it stick. Administrative fines run to EUR 15 million or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. Two more dates follow this one: generative systems already on the market must reach marking conformity by 2 December 2026, and a detection interoperability solution is due by 2 February 2027.
Where does that leave an advertiser?
The platforms have been signing up to the voluntary framework that supports compliance. Google signed the Code of Practice on Transparency of AI-Generated Content on 24 July, cautioning at the same time that added regulatory complexity in an unsettled technical field could work against European competitiveness. Meta signed on 28 July, five days before the obligations bound, having refused the separate General-Purpose AI Code in July 2025 over what it called legal uncertainties.
Signing a code is not the same as shipping a control. Meta's announcement describes principles and forums, and no corresponding labeling setting has appeared in Ads Manager. An advertiser reading the rule today finds an obligation with a named icon, a specified placement, and a fine attached, and no button in the tool where the creative is uploaded.
That leaves the practical burden with whoever publishes. If AI generated the copy or the imagery in an ad running to EU users, the visible label is something to place in the asset and on the page it lands on, because there is currently nowhere else to place it.
What to do about it
Work out whether you are in scope
The trigger is publishing AI-generated content to people in the EU, not being an EU company. A US store running Meta or Google campaigns into European markets, or serving AI-written product copy to EU visitors, is a deployer under the definition. Establish that first, because everything downstream depends on the answer.
Inventory what you already publish as AI-generated
Separate the deepfake and AI-generated-published-text categories the article names from the broader pile of AI-assisted work, and list where each one appears: ad creative, landing pages, blog posts, product imagery. You cannot label what you have not catalogued, and the catalogue is also what tells you how large the exposure actually is.