Insights
AI Watermarks Are Here: What the EU AI Act Means for Your Content
On 2 August 2026, Article 50 of the EU AI Act began to apply. If your company uses AI to research, write, edit or produce content, these are the first EU-wide transparency rules that directly touch that process.
I think this is where many marketing teams are getting the story slightly wrong.
The new rules do not mean that every article touched by AI needs a warning label. They also cover much more than watermarks. Article 50 deals with chatbots, machine-readable marking, deepfakes, emotion recognition and certain AI-generated text.
For most B2B content teams, the decisive questions are quite practical: what did the AI produce, what did a human do afterwards, and who took responsibility before the content went live?
What Article 50 Actually Requires
The official text of Article 50 sets out several transparency obligations. They apply to different actors and different uses of AI, which explains some of the confusion around the new rules.
AI chatbots must identify themselves
AI systems designed to interact directly with people must make it clear that the user is dealing with an AI system, unless this is already obvious from the context.
For a marketing team, this can apply to customer-service chatbots, website assistants and conversational tools used to answer product questions or qualify leads.
If a visitor can reasonably believe that they are speaking to a person, the system needs to make its artificial nature clear.
AI-generated content must be machine-readable
Providers of AI systems that generate or manipulate text, audio, images or video must ensure that their outputs can be detected as artificially generated or manipulated.
This is the technical watermarking part of Article 50. The obligation sits primarily with the provider of the AI system, such as OpenAI, Google or the developer of an image, audio or video generation tool.
The law also includes an important qualification. The marking obligation does not apply when an AI system performs a standard editing function or does not substantially alter the input or its meaning.
That detail changes the conversation.
Using AI to correct grammar, improve the structure of a paragraph or suggest several headlines is different from generating an entire article and publishing the output with minimal intervention.
AI-assisted content and AI-generated content are not automatically the same thing.
Deepfakes need visible disclosure
Companies using AI to generate or manipulate images, audio or video that could falsely appear authentic must disclose that the content is artificial.
This matters for synthetic spokesperson videos, cloned voices, realistic product imagery and fictional scenes presented in a way that could be mistaken for reality.
Creative, artistic, satirical and fictional works receive more flexible treatment. The disclosure must still exist, although it should be presented in a way that does not interfere with the experience of the work.
Certain AI-generated text must also be disclosed
Article 50 covers AI-generated or manipulated text published for the purpose of informing the public about matters of public interest.
This is the part that makes publishers and content teams nervous. It is also the part with the most relevant exception.
The Exception Most Content Teams Already Meet
The public-interest text disclosure requirement does not apply when two conditions are met:
- The content has undergone human review or editorial control.
- A natural or legal person holds editorial responsibility for its publication.
Both conditions matter.
A byline alone is weak evidence of editorial responsibility if the article went directly from a language model to the website. Someone must have reviewed what the model produced, checked the claims and accepted responsibility for the final publication.
In a real editorial workflow, that usually means:
- Reading and editing the complete draft
- Checking facts, quotations and sources
- Correcting inaccurate or misleading claims
- Approving the final version
- Recording who was responsible for publication
The distinction I would make is simple: using AI does not remove editorial responsibility. It makes that responsibility more important.
AI-assisted research, first drafts, structural suggestions and headline ideas do not automatically trigger a disclosure requirement. The nature of the output, the purpose of the publication and the level of editorial control all matter.
A workflow that sends raw AI output directly to a live website creates a different situation. If that text is intended to inform the public about a matter of public interest, disclosure may be required.
Scale is not the legal test. A single publication can fall within Article 50. Publishing hundreds of unreviewed pages makes the problem larger because the same weak process is being repeated.
Does “Public Interest” Include Your Company Blog?
There is no useful reason to classify every company blog as public-interest publishing. It would also be risky to assume that commercial content can never fall within that category.
“Matters of public interest” can include information concerning public policy, health, safety, finance, rights, the environment, elections and other subjects that influence public understanding or decision-making.
A guide to schema markup or a comparison between SEO and AEO is less likely to fall into this category than an article offering medical guidance, financial advice or an interpretation of public regulation.
Context still matters.
The European Commission published its final guidelines on Article 50 on 20 July 2026. The guidelines clarify the scope of the obligations, relevant definitions and practical exceptions.
The Commission has also approved a voluntary Code of Practice on Transparency of AI-generated Content. It gives providers and deployers a recognised framework for demonstrating compliance with the marking and disclosure rules.
My reading for a normal B2B content operation is straightforward: editorial control should be part of the system.
Keep a human involved before publication. Document that the review happened. Make clear who owns the final decision. Escalate uncertain cases when the content deals with public-interest or regulated subjects.
This is a stronger approach than treating human review as a legal escape hatch added at the end of an automated process.
What This Changes for Content Marketing
The immediate effect of Article 50 is not a new badge on every blog post. It is a clearer division of responsibility between AI providers and the companies using their systems.
The provider is generally responsible for the technical marking of generated outputs. The company publishing a deepfake or certain public-interest text is responsible for the disclosure visible to the public.
For content teams, the real shift is operational.
If AI is now part of the production workflow, editorial review can no longer live as an informal assumption. Teams need to know:
- Where AI enters the process
- Which parts of the content it generates or changes
- Who checks the result
- Who approves publication
- Whether synthetic media requires separate disclosure
None of this needs to become a bureaucratic machine. A clear workflow and a meaningful sign-off step can cover much of it.
Why This Matters for AEO and GEO
The EU AI Act does not establish AI watermarks as an AEO ranking factor. There is no public evidence that adding a label or author byline directly makes an answer engine more likely to cite a page.
The connection with AEO happens at the editorial level.
Google’s guidance on generative AI content focuses on accuracy, quality, relevance and context for users. It also warns that generating large numbers of pages without adding value may violate its policy on scaled content abuse.
Google’s newer guidance for generative search goes further. It recommends original, useful and non-commodity content for visibility in AI Overviews and AI Mode. It also confirms that established SEO practices remain relevant because these AI features rely on Google’s search index and core ranking systems.
That makes sense.
AI answer engines have little use for another generic page repeating what already exists. They need sources containing something worth retrieving: a clear explanation, original experience, reliable evidence or a point of view that adds meaning.
Human review does not create authority by itself. It creates the opportunity to remove errors, question weak claims and add what the model could not know.
This is where compliance, SEO and AEO begin to meet. They all benefit from content with a clear source, real responsibility and something specific to say.
We have already seen the other side of this pattern in what AI-generated content is doing to brand identity. As more companies use the same systems to produce the same type of content, sameness becomes visible very quickly.
The advantage belongs to companies that use AI as part of a thinking process, rather than as a replacement for one.
A Practical Checklist for B2B Marketing Teams
Five habits cover the most important areas.
1. Record how AI was used
Was the system used for research, editing, translation, image generation or the complete first draft?
This helps distinguish assistance from generation or substantial manipulation.
2. Assign a real editorial owner
Every published piece should have a person or organisation responsible for approving it.
A byline should mean that someone read the article, challenged it and decided it was ready to represent the company.
3. Review the final content
Check facts, sources, quotations, legal claims and any conclusions introduced by the AI system.
Review the version the public will see. Approving a prompt or outline is not the same thing.
4. Keep evidence of the review
A dated revision history, CMS approval, editorial log or documented sign-off can show that the process existed.
You do not need to build a miniature European Commission inside the marketing department. You need a process that is clear enough to repeat and explain.
5. Treat synthetic media separately
AI spokesperson videos, cloned voices, realistic synthetic photography and deepfake-style material may require disclosure even when the accompanying text has gone through editorial review.
The editorial exception for public-interest text does not remove the separate rules for synthetic media.
Do You Need a “Written by AI” Badge?
For a conventional B2B article that has been genuinely reviewed and carries clear editorial responsibility, usually no.
Article 50 does not create a general obligation to label every article that involved AI somewhere in its production.
It does require companies to understand what their systems produced and to maintain real control over what they publish.
For me, this is the most useful way to understand the regulation:
AI can be part of the process. A person still has to own the result.
That is also the kind of content operation that has a better chance of producing something people want to read, trust and cite.
The Short Version
Article 50 of the EU AI Act applies from 2 August 2026. It introduces transparency rules covering AI interactions, machine-readable marking, deepfakes and certain AI-generated text published on matters of public interest.
Most human-reviewed B2B blog content will not require an AI disclosure when it has gone through genuine editorial control and a person or organisation holds responsibility for its publication.
Companies should distinguish between AI assistance, AI generation and substantial AI manipulation. They should also be able to show how content was reviewed before publication.
The law sets a minimum. Good editorial work should already go further.
This article provides general information and does not constitute legal advice. Companies publishing high volumes of AI-generated content, operating across several EU markets or covering regulated and public-interest subjects should seek advice on their specific workflows.
Frequently Asked Questions
Do I have to label AI-assisted blog posts as AI-generated?
Usually, no. The use of AI somewhere in the production process does not automatically create a disclosure requirement.
For AI-generated or manipulated text concerning matters of public interest, Article 50 provides an exception when the publication has undergone human review or editorial control and a person or organisation holds editorial responsibility.
When did Article 50 take effect?
The transparency obligations began to apply on 2 August 2026.
Certain AI systems placed on the market before that date have a transitional compliance period until 2 December 2026.
Does Article 50 apply to companies outside the EU?
It can.
The AI Act covers providers placing AI systems on the EU market, deployers established or located in the EU, and certain providers or deployers outside the EU when the output produced by the system is used within the EU.
The precise application depends on the company’s role, location and use of the AI system.
Does machine-readable watermarking apply to marketing text?
Article 50(2) covers AI systems generating or manipulating text as well as audio, images and video. The technical marking obligation sits primarily with the provider of the AI system.
A separate obligation under Article 50(4) covers companies publishing AI-generated or manipulated text for the purpose of informing the public about matters of public interest.
Does human review always remove the need for disclosure?
No. The editorial exception discussed here applies specifically to AI-generated or manipulated public-interest text.
Deepfakes and other forms of synthetic media are covered by separate disclosure rules. Human review does not automatically remove those obligations.
Primary Sources
Put it into practice
Know what matters now. Fix what matters first.
No spam. No commitment. Just a straight answer.