AI Labeling Requirements: What Marketing Teams Must Do

The AI labelling obligation is an EU requirement under Article 50 of the AI Act (EU AI Act) that takes effect on 2 August 2026, requiring companies to label certain AI-generated content as such. Not every use of AI in marketing requires a label – the obligation only applies when synthetic content could be mistaken for real people, places or events, or when users interact with an AI system without knowing it. Whether AI labelling is necessary depends on the content type, the risk of confusion and the degree of editorial control. This article provides a three-tier traffic-light framework for rapid self-assessment by marketing teams – and shows where the line runs between obligation, recommendation and all-clear.

What the AI transparency obligation under the EU AI Act regulates

The EU AI Act entered into force on 1 August 2024 and takes effect in stages. Article 50 governs the transparency obligations for providers and deployers of AI systems – these become binding on 2 August 2026. The provision is not aimed solely at technology providers; it applies to any company that publishes AI-generated content or deploys AI systems in customer-facing contexts. The AI transparency obligation is not a blanket prohibition but a differentiated set of rules with clearly defined exceptions.

Most companies underestimate how much AI already runs quietly through their daily operations, long before anyone has written down a single rule for it. Shadow AI is not a footnote in the risk report – it is the risk, and it grows every time an employee pastes sensitive data into a tool nobody approved. Whether the use of AI stands on GDPR-compliant ground is not a question of good intentions; it is a question of whether audits, policies and secure integrations exist that can actually hold the promise of compliance. How AI governance and compliance are put on solid legal footing is worth examining for anyone who wants to know not only that their AI usage is safe, but why.

Which content falls under the labelling obligation

The regulation targets three categories that are directly relevant to marketing:

• Deepfakes: AI-generated or AI-manipulated image, video and audio content that bears a deceptive resemblance to real people, places or events. A photorealistic portrait of a non-existent person in a LinkedIn campaign falls into this category.
• AI chatbots in user-facing interactions: Any system that interacts with people and could create the impression of being human must be labelled as an AI system – regardless of the quality of its responses.
• Synthetic content with a risk of confusion: Texts, images or videos that, without labelling, could be taken for authentic, human-created content and are published to an audience.

When the labelling obligation does not apply – the exceptions

The AI transparency exceptions are the decisive lever for marketing teams. Article 50 explicitly provides that the labelling obligation does not apply when:

• Editorial control is exercised by a natural or legal person who takes responsibility for the content. A blog article that a human reviews, revises and publishes under their name remains exempt from labelling – even if the draft originated from a language model.
• Obviously artistic or fictional content is involved. A stylised illustration in flat-design style does not trigger any risk of confusion.
• Purely assistive edits such as colour correction, background removal or format adjustments are made that do not substantially alter the content.
• Internal use without external publication takes place – for example, AI-generated drafts for internal presentations.

The AI marketing traffic light – self-assessment in three tiers

The regulation demands differentiation, not panic. The following AI marketing traffic light translates the legal categories into a self-assessment that works without legal training. The logic: the higher the risk of confusion with real content and the lower the degree of human control, the more urgent the need for action.

Green – no labelling required

A blog article whose draft is produced by a language model and then reviewed, rewritten and approved by an editor falls under the editorial-control exception. The same applies to newsletter copy with AI assistance followed by human editing, internal presentations, and social-media graphics in a recognisably stylised design. An explainer video with a synthetic voice that sounds recognisably artificial and cannot be attributed to a real person also remains in the green zone. The common thread: either human responsibility is in place, or the content poses no risk of confusion.

Yellow – case-by-case review recommended

AI-generated product images that contain realistic elements – for example, a photorealistic kitchen scene with an AI-extended background – occupy a grey area. The same applies to video content in which synthetic and real elements merge. The recommendation for the yellow zone: when in doubt, label voluntarily. The cost of labelling is low; the cost of non-compliance is not.

Red – immediate action required

Photorealistic AI images in campaigns that depict people who do not exist or resemble real individuals clearly fall under the labelling obligation. Synthetic voices that imitate real people are deepfakes within the meaning of the regulation. And an AI chatbot on a website that communicates with customers without disclosure will be in breach of applicable law from 2 August 2026. This is where immediate action is needed – not at some point in the future, but within the remaining 13 days.

Fines and liability risks for violations of the AI Act

Article 99 of the AI Act provides a three-tier sanctions system calibrated to the severity of the violation. The fines are deliberately set high – comparable to the GDPR, but significantly exceeding it at the upper end. For marketing teams, the middle tier is the relevant one: violations of the transparency obligations under Article 50 can be sanctioned with up to EUR 15 million or 3 % of global annual turnover – whichever amount is higher.

Worked example: A company with EUR 200 million in annual turnover risks a maximum of EUR 6 million for a transparency violation (3 % of turnover). Since this amount is below the flat threshold of EUR 15 million, the turnover-based calculation applies. For a company with EUR 600 million in turnover, the figure would be EUR 18 million – in that case, the flat threshold no longer acts as a cap; the turnover share prevails. These numbers are not theoretical. They are the framework within which national supervisory authorities can impose fines from August 2026 onwards.

How labelling works in practice – technically and visually

The AI Act does not prescribe an exact wording. It requires "clear and distinguishable information" at the point of first interaction or perception. This means: the label must appear where the content is consumed – not buried in the legal notice, not in the terms and conditions, not in a separate disclosure document.

Visible labelling for humans

For AI content labelling in marketing, short, unambiguous statements placed directly alongside the content are appropriate: "Created with AI", "AI-generated image", "This text was produced with the assistance of an AI system". Placement is directly at the content – for images as an overlay or caption, for chatbots as a notice before the first interaction. What matters is not the exact wording but recognisability for the average user.

Machine-readable markers and metadata

In addition to visible labelling, the regulation requires machine-readable markers. The C2PA standard (Coalition for Content Provenance and Authenticity) has established itself as the technical framework. Digital watermarks and metadata document the provenance of content and must not be removed – deleting metadata constitutes a violation of the regulation. Platforms such as YouTube, LinkedIn, Meta and TikTok have introduced their own labelling rules in parallel, some of which go beyond the EU requirements. Anyone publishing across multiple channels needs to be aware of the platform-specific additional rules.

AI marketing checklist – five steps to compliance

Companies that publish AI-generated content need documented processes – not as a bureaucratic box-ticking exercise, but as evidence for supervisory authorities. Anyone who can demonstrate in an enforcement scenario that responsibilities are defined, approvals are documented and staff are trained will be in a stronger position than a company with no structure at all. The following five steps represent the minimum for AI marketing compliance.

A documented AI policy makes responsibilities and approval workflows plannable. Companies that prefer not to handle implementation in-house can develop it with a specialised communications agency such as Crispy Content®.

The EU Code of Practice from June 2026 – what it changes for marketing teams

On 10 June 2026, the European Commission published the final "Code of Practice on Transparency of AI-Generated Content". Participation in the Code is voluntary – the underlying obligations from Article 50 remain binding regardless. The Code specifies what the regulation formulates in abstract terms and provides guidance to companies along the entire AI value chain.

The Code of Practice structures responsibilities along the chain Provider → Deployer → Platform. It recommends concrete measures on watermarking, metadata and visible labelling – without mandating them. For marketing teams, the relevance lies in evidential value: anyone who follows the Code and documents their adherence can demonstrate in a dispute that reasonable compliance efforts were undertaken. That is not a free pass, but it is an argument. The Code does not replace legal advice; it supplements it with a practical framework.

Trends and outlook – how the AI transparency obligation will evolve

The regulatory landscape is dynamic. Further specifications through implementing acts and national supervisory authorities are forthcoming – the regulation itself is the framework, not the final word. Three developments deserve particular attention:

• Case law will clarify the exceptions: Where exactly the line falls between "editorial control" and mere sign-off will need to be determined by courts. The first proceedings are likely to provide guidance in 2027.
• Platforms are tightening rules in parallel: YouTube, Meta and TikTok have introduced their own labelling rules, some of which are stricter than the EU requirements. Anyone who reads only the regulation overlooks the platform layer.
• International convergence: South Korea's AI Basic Act and India's IT Rules 2026 are increasing global pressure. Companies with international reach will not be able to confine themselves to a single jurisdiction.

The AI literacy obligation under Article 4, in effect since February 2025, is the quiet prerequisite for everything that follows: anyone who does not know what AI can do and where it is deployed can neither label correctly nor assess risks.

Traffic-light use cases in everyday marketing

Theory becomes tangible when it meets concrete marketing activities. The following table shows eight typical scenarios and their classification under the traffic-light framework:

The pattern is clear: text with human responsibility is green. Photorealistic images depicting people are red. In between lies a narrow yellow band that requires case-by-case assessment.

Differentiation, not panic – what marketing decision-makers should prioritise now

Most editorially controlled marketing texts fall under green – they require no labelling as long as a responsible person reviews and approves the content. The need for action centres primarily on two categories: photorealistic AI images depicting people and AI chatbots without a transparency notice. Anyone who addresses these two issues in the remaining days before 2 August has dealt with the most critical gaps. Documented processes and clear responsibilities are the key – not because the regulation loves bureaucracy, but because in an enforcement scenario, the company that can show it thought before it published will always be in a stronger position.

Sources

Haufe Online Redaktion (2026): Kennzeichnungspflicht für KI-Inhalte gilt ab August 2026. URL: https://www.haufe.de/recht/kanzleimanagement/kennzeichnungspflicht-fuer-ki-inhalte-gilt-ab-august-2026_222_681220.html (accessed 20.07.2026).

IHK Köln (2026): KI-Verordnung: Neue Transparenzpflichten ab August 2026. URL: https://www.ihk.de/koeln/hauptnavigation/digitalisierung-und-innovation/digitalisierung/transparenzpflichten-nach-der-ki-verordnung-7100068 (accessed 20.07.2026).

Radikant GmbH / Christian Wonner (2026): Was der EU AI Act für Marketing-Teams bedeutet – KI-Kennzeichnung wird Pflicht. URL: https://www.radikant.com/wissen/ki-kennzeichnung-wird-pflicht-bedeutung-fuer-marketing-teams (accessed 20.07.2026).

Vertretung der Europäischen Kommission in Deutschland (2026): Kennzeichnung von KI-generierten Inhalten: EU-Kommission veröffentlicht Verhaltenskodex. URL: https://germany.representation.ec.europa.eu/nachrichten-und-veranstaltungen/pressemitteilungen/kennzeichnung-von-ki-generierten-inhalten-eu-kommission-veroffentlicht-verhaltenskodex-2026-06-10_de (accessed 20.07.2026).

Secjur (2026): EU AI Act im Mai 2026: Pflichten, Risikoklassen und Stand. URL: https://www.secjur.com/blog/eu-ai-act (accessed 20.07.2026).

Europäische Kommission (2026): Code of Practice on Transparency of AI-Generated Content. URL: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content (accessed 20.07.2026).

Europäische Kommission (2026): Commission publishes Code of Practice on marking and labelling of AI-generated content. URL: https://digital-strategy.ec.europa.eu/en/news/commission-publishes-code-practice-marking-and-labelling-ai-generated-content (accessed 20.07.2026).