On August 2, 2026, the transparency provisions of the European Union's AI Act took effect. If your marketing team uses AI to write blog posts, ad copy, product descriptions, or customer emails, this law may apply to you. It applies because you publish content, not because you built an AI model.
The confusion in the market right now is understandable. The law creates four separate obligations that apply to different groups. Most companies are scrambling to figure out which rules apply to them and what they need to do. The core concern is practical: companies are worried about the content they produce, not what AI search engines cite about them. They want to know if the blog posts, ad copy, and product descriptions their teams create with AI need to be labeled.
Here is what the law actually says, who it applies to, and what you should do.
The law applies to two groups: providers and deployers
The EU AI Act Article 50 draws a line between providers (companies that build AI systems) and deployers (companies that use them). The obligations are different for each.
Providers are companies like Anthropic, OpenAI, and Google. If you build a generative AI system that produces text, images, audio, or video, you must mark your outputs in a machine-readable format so they can be detected as artificially generated. You must also ensure that chatbots and virtual assistants disclose their AI nature to users.
Deployers are companies that use AI systems. If you use AI to generate content that you publish, this is where your obligations begin.
Four obligations, only two likely apply to your business
Article 50 creates four transparency rules. Two are aimed at providers. Two affect deployers.
Obligation 1 (providers): Chatbot disclosure
If you operate a customer-facing chatbot or virtual assistant, users must be informed they are interacting with AI. This applies at the start of every interaction, not buried in terms and conditions.
Obligation 2 (providers): Machine-readable marking
Generative AI systems must mark outputs as artificially generated in a machine-readable format. This is the watermarking requirement that prompted Anthropic's August 11 announcement. The technical standards are still being finalized through a voluntary Code of Practice, but a standardized EU label is being developed.
Obligation 3 (deployers): Deepfake disclosure
If you use AI to create or manipulate image, audio, or video content that resembles real people, places, or events, you must disclose that the content is artificially generated. There is an exception for clearly artistic, creative, satirical, or fictional work, where the disclosure just needs to note that AI-generated content exists somewhere in the piece.
Obligation 4 (deployers): AI-generated text on matters of public interest
This is the one that has companies scrambling. If you publish AI-generated text with the purpose of informing the public on matters of public interest, you must disclose that the text is AI-generated.
The exception that matters most
Article 50(4) includes an exception that most responsible publishers will rely on. The text labeling requirement does not apply when two conditions are both met:
- The AI-generated content has undergone a process of human review or editorial control, and
- A natural or legal person holds editorial responsibility for the publication.
The law is specific about what counts. The review must be "substantive and not limited to superficial matters or cursory approval." A quick proofread does not qualify. A genuine editorial review where a human checks facts, evaluates claims, revises language, and takes responsibility for the published version does qualify.
This means companies that already have strong editorial processes are in better shape than they might think. If a human reviews, edits, and stands behind the content, the labeling requirement likely does not apply.
What counts as "matters of public interest"?
This is the question companies are struggling with most. The law does not define the term precisely. The European Commission's draft Guidelines indicate that the trigger turns on the publisher's purpose, not merely the subject matter.
In practice, this means:
- Likely covered: News articles, investigative reports, analysis pieces about public affairs, health information, consumer protection content, educational content about public issues
- Likely not covered: Product descriptions, advertising copy, promotional blog posts about your own products, internal communications, customer service responses
The gray area is brand-published content that blurs the line. A company blog post about industry trends could be "matters of public interest." A product comparison guide might not be. A thought leadership piece on regulatory changes probably is. There is no bright line yet.
What about marketing content specifically?
Most marketing content falls outside the "matters of public interest" scope. Product pages, ad copy, email campaigns, and promotional materials are commercial speech, not public interest content.
But there is a catch. If your marketing team publishes content designed to inform or educate (guides, explainers, industry analysis), and that content could influence public understanding of an issue, it may cross into public interest territory. Companies publishing educational content at scale face the highest compliance risk.
The practical compliance checklist
If your team uses AI to produce content, here is what you should do now:
Audit your content pipeline. Identify where AI is used in your content production. Map every workflow: drafting, research, editing, translation, summarization. You need to know which published content involved AI.
Classify your content. Sort your published content into three buckets: clearly commercial (product pages, ads), clearly public interest (news, analysis, education), and unclear (industry blog posts, thought leadership). The middle bucket needs the most attention.
Document your editorial process. For content that could be public interest, document that human review occurred. Keep records of who reviewed, what they checked, and when. The law requires the review to be substantive. If you cannot prove it happened, you cannot claim the exception.
Label what needs labeling. For content that is AI-generated and public interest, and has not undergone substantive human review, add a disclosure. The EU is developing a standardized "AI" label, but until it is finalized, clear text disclosure is the safe approach.
Review your chatbot and virtual assistant. If you deploy any customer-facing AI assistant, verify that it discloses its AI nature clearly at the start of every interaction.
Key dates
- February 2, 2025: Prohibitions on unacceptable AI practices took effect
- August 2, 2026: Article 50 transparency obligations took effect
- December 2, 2026: Machine-readable marking deadline for AI systems already on the market before August 2, 2026
- June 2026 (expected): Code of Practice on AI-generated content finalized
The bottom line
Most companies do not need to panic. If you have a genuine editorial review process where humans check, revise, and stand behind published content, you likely qualify for the human-review exception. The companies at risk are those publishing AI-generated content at scale without meaningful human oversight.
The law is not trying to stop businesses from using AI. It is trying to ensure that when AI generates content published to the public, someone is accountable for accuracy. For companies that already take editorial responsibility seriously, compliance is a documentation problem, not a fundamental change in how you work.
For companies that have been publishing raw AI output with no review, August 2 was a wake-up call.