EU AI Act Article 50: an advertising production guide

AI can expand what a brand campaign can look like. It also changes the questions a producer needs to answer before an asset goes live: what was generated, who controlled the AI system, and what should the audience be told?

The EU AI Act’s Article 50 transparency rules apply from 2 August 2026. For advertisers, the useful distinction is between the AI provider’s technical marking duties and the deployer’s audience-facing disclosure duties. A limited transition to 2 December 2026 concerns Article 50(2) marking for systems placed on the market before 2 August; it is not a general postponement for campaigns. See the European Commission’s Article 50 FAQ.

AI-generated imagery from Martinelli’s Valentine’s campaign, produced by PHOCUS. View the project and credits.

Who is responsible: the tool, the agency or the production company?

A provider develops an AI system, or has it developed, and brings it to market or into use under its own name. A deployer uses an AI system under its authority in a professional context. A production company can therefore be a deployer; the role does not automatically belong to whoever buys the media or publishes the final film. The Commission also explains that a company can remain the deployer when contractors operate the system on its behalf and under its control. Source: Commission FAQ.

For a commission, map the actual working arrangement before assigning tasks. Identify who selects the tools, directs their use, approves the generated material and releases each version. Put named owners against these decisions in the production brief. Contractual responsibilities should reflect that workflow rather than assume that outsourcing production settles the legal question.

Technical marking and visible disclosure are different

Article 50(2) addresses providers of systems that generate synthetic audio, images, video or text. It requires machine-readable marking and detectability, subject to the provision’s qualifications, including an exception for assistive standard editing or changes that do not substantially alter the input or its meaning.

Article 50(4) addresses deployers when generated or manipulated image, audio or video constitutes a deepfake. A deepfake is defined by its resemblance to existing people, objects, places, entities or events and its false appearance of authenticity or truthfulness. The same paragraph separately covers AI-generated or manipulated text published to inform the public on matters of public interest, with an exception involving human review or editorial control and editorial responsibility. These are distinct tests, not a blanket rule that every use of AI in advertising needs the same label. Read Article 50 and the statutory definitions.

For a production team, this means keeping two separate checks in the delivery process: the technical provenance of the asset and the disclosure that a viewer may need to see or hear. Do not treat an export’s metadata as the creative team’s answer to both questions.

Assess the finished campaign in context

The Commission’s guidance considers both the content and its deployment context, including what the audience would expect to be authentic. Evidently creative or fictional works have an adapted disclosure provision; calling an advertisement “creative” is not by itself an assessment. Where disclosure is required, Article 50(5) calls for clear, distinguishable and accessible information by first exposure. Source: Commission guidelines on Article 50.

Review the intended impression as well as the technique. Does the image suggest a product performed a real test? Does a voice imply a real person endorsed the brand? Could a location or event be understood as documentary evidence? These are useful questions for a campaign review, not automatic legal classifications. A speculative treatment and a finished paid advertisement may need different assessments.

A second image from the Martinelli AI campaign. Product detail and the audience’s likely interpretation both deserve attention during creative review.

A practical workflow from brief to delivery

The following is a recommended production workflow. It is not a claim that Article 50 mandates every document or review step below.

1. Mark AI use in the treatment

Separate live-action footage, conventional post-production and generative work. Describe the planned use of synthetic people, voices, products and locations. Include the target territories and placements so the review concerns the campaign you will actually distribute.

2. Agree responsibilities and permissions

Name the production, agency and brand contacts responsible for AI-use records, rights review and final release. Check tool terms and the permissions needed for supplied photographs, performances, likenesses and voices. Confirm the agreed uses in the relevant licences and contracts.

3. Keep an asset record during production

For material intended for delivery, record the tool, relevant version or generation date, source assets and subsequent edits. Keep available provenance information with the master files. A simple asset register makes later amendments easier than reconstructing the process from scattered messages.

4. Review each placement before export

Check the final film, cutdowns, stills and localised versions. If disclosure is needed, agree wording and placement with the responsible reviewers. Test legibility, safe areas, duration and accessibility on the actual format. A label that works on a landscape master may disappear in a vertical crop.

5. Hand over a usable release package

Deliver approved assets with an AI-use summary, relevant permissions, disclosure instructions and named approval contacts. Keep a reference copy of the released versions. Give the media team enough information to avoid removing a necessary disclosure when adapting an asset.

Briefing an AI or hybrid production with PHOCUS

A useful enquiry includes the creative idea, product references, planned deliverables, markets, launch date and whether you are considering live action, AI or a combination. Flag any real people, voices or performance claims early so they can be discussed before the approach is locked.

Explore our Martinelli AI project, then send PHOCUS your campaign brief to discuss the production approach and scope.

Updated 9 September 2026. This article is a production planning guide, not legal advice. Transparency requirements do not replace the separate review of advertising claims, intellectual property, privacy or personality rights.