Europe’s AI transparency rules have been live since 2 August 2026, and a new Austrian guide gives marketers a clearer answer on who owns the label: the agency running the AI tool usually does, not the brand paying for the campaign.
The default: agencies carry the deployer role
Published 16 September 2026 by IAB Austria’s AI working group and act legal Austria, the five-page guide walks through Article 50 of the EU AI Act. The core distinction is between a provider – usually the software maker – and a deployer – the party using the tool in production and deciding how. For most campaign creative, the producing agency is the deployer and must assess whether a label is needed and add it technically.
Clients stay outside that duty unless they step into the AI use themselves: prescribing the tool, handing over prompt instructions, setting parameters or using their own AI systems. The guide says merely knowing AI was used is not enough to make a brand a deployer, but repeated prompt-level correction rounds can blur the line.
What actually triggers a label
Not every AI-assisted asset must be labelled. The guide’s exemptions are the most useful starting point: abstract AI graphics, cartoon characters, obviously fictional depictions and classic technical edits such as noise reduction, colour correction and retouching are outside the duty.
- Realistic media that could deceive: deepfakes, cloned voices, fake testimonials, manipulated places or events.
- Public-interest AI text published without prior editorial control.
- Chatbots and support tools that interact directly with users.
Pre-August 2026 content does not need retroactive labels, but the duty applies when it is newly published or served again – a detail worth reviewing for always-on creative.
Visible means visible
A watermark won’t save you. The guide says SynthID, C2PA metadata or a note in the imprint and terms and conditions are not sufficient. The label must be clearly visible in the asset itself, or audible for audio, and the EU’s free icon set can help: “AI GENERATED” for fully synthetic content and “AI MODIFIED” for partially altered work.
That distinction matters because Google applies imperceptible signals automatically to AI images and videos inside its ads tools. Under the Austrian reading, those machine-readable marks alone do not meet a user-visible disclosure.
Turn it into a contract process
The practical takeaway is operational, not just legal. The guide wants agencies, clients, publishers and tech providers to settle three questions in bilateral or multilateral contracts: who determines whether content must be labelled, who applies and checks the label, and who is liable if it is missing.
Breaches can reach €15 million or 3% of worldwide annual group turnover, whichever is higher, and can also trigger unfair competition claims under Austria’s UWG. Since labelling too much has a documented cost – one NYU Stern study cited in the IAB framework found AI disclosure cut ad click-through rate by 31.5% – marketers need a disciplined process, not over-labelling.
What marketers should do now
Before the next launch, map every AI-assisted asset against the three triggers and document the decision. Agree label ownership in contracts, and keep the visible EU icon in the creative, not just in an ad account setting. If you run a white-label AI tool under your own brand, check whether you have crossed into provider status too.
Agency, brand and publisher may each carry different roles under the AI Act, platform rules and national competition law. The contract is where those roles should align before an asset goes live.
Source: PPC Land
