Artificial intelligence is already part of the daily operations of many companies. Marketing teams use AI for research, campaigns, and automation. In sales, AI can support lead qualification, while in customer support, chatbots and virtual assistants can answer customer questions.
However, as the use of these technologies expands, new responsibilities also emerge. Starting on 2 August 2026, the transparency obligations under Article 50 of the AI Act apply. Their purpose is to help users identify situations in which they are interacting with an AI system or are exposed to content created or manipulated using artificial intelligence.
This does not mean that every article, image, or presentation created with the help of AI must automatically carry a label. The rules vary depending on the type of system, the content, how it is presented, and the company’s role. In this article, we explain what the AI Act requires, when transparency is necessary, and what marketing teams should verify in practice.
This article is for informational purposes only and does not replace legal advice tailored to your company’s specific circumstances.
The AI Act, the commonly used name for Regulation (EU) 2024/1689, is the European legal framework establishing rules for the development, placing on the market, and use of artificial intelligence systems.
The regulation aims to support the safe and trustworthy use of artificial intelligence while protecting health, safety, and fundamental rights. Its approach is based on the level of risk associated with different uses of AI.
Depending on the level of risk, AI systems and applications are treated differently:
For most marketing teams, the most visible changes introduced in August 2026 relate to transparency.
Article 50 of the AI Act introduces obligations for certain interactive and generative AI systems, as well as for content that may be mistaken for authentic content.
The purpose is not to label every piece of content that AI has helped create. The rules primarily target situations in which a lack of transparency could mislead the public, facilitate impersonation, or undermine trust in the information people consume.
For companies, relevant obligations may arise when:
The AI Act makes an important distinction between providers of AI systems and the organisations that use them in their professional activities.
A provider is, in principle, an organisation that develops an AI system or places it on the market under its own name or brand.
For systems that generate content, providers must ensure that outputs can be identified as artificially generated or manipulated. This may involve machine-readable markings designed to allow the content to be detected through technical means.
A deployer is a company or organisation that uses an AI system as part of its professional activity.
An agency using an AI platform to generate campaign images, a company installing a chatbot on its website, or an organisation publishing AI-generated material may be considered a deployer. This is the role in which most marketing teams will find themselves.
Depending on the use case, the deployer may be responsible for informing the public or visibly labelling the material.
In some cases, the same company may have multiple roles at the same time, depending on the system it develops and how it uses it.
When a person interacts directly with an AI system, the provider must ensure that the person is informed that they are communicating with an artificial system.
For a chatbot installed on a website, the message could be simple:
“You are chatting with an AI-powered virtual assistant.”
The information should be presented clearly and at the appropriate time, rather than being hidden in a lengthy terms and conditions page.
The regulation includes an exception for situations in which it is obvious, from the perspective of a reasonably well-informed, observant, and circumspect person, that the interaction is taking place with an AI system.
However, the Commission’s guidelines indicate that this exception should be interpreted narrowly. A company should not automatically assume that users will “figure it out.”
This obligation may be relevant not only for customer support, but also for marketing and sales. For example, a chatbot may recommend services, qualify leads, or schedule a conversation with a consultant.
A deepfake is audio, video, or visual content generated or manipulated using AI that resembles real people, objects, places, or events and may falsely be perceived as authentic.
This category may include:
In such situations, the deployer must disclose that the material has been artificially generated or manipulated.
However, not every image created with AI is a deepfake. A clearly stylised illustration, an abstract graphic, or an image that does not claim to represent reality will not automatically be treated in the same way as a photograph of a real person.
The practical question for a marketing team is:
Could a reasonable person believe that this material represents a real person, statement, or event?
When the answer is yes, the material requires careful assessment and, where applicable, clear labelling.
Article 50 also includes a rule for AI-generated or AI-manipulated text published for the purpose of informing the public about matters of public interest.
Depending on the context, this may include topics such as:
This does not mean that every blog article created with the help of AI must be labelled.
The rule includes an important exception: labelling is not required when the text has undergone human review or editorial control and a natural or legal person assumes editorial responsibility for its publication.
An article does not automatically become “AI content” simply because a tool was used for brainstorming, structuring, rewriting, or proofreading.
What matters is how the information was verified and who takes responsibility for the final result.
In practice, a company should be able to demonstrate that:
Therefore, a genuine editorial process is more valuable than simply adding a label.
People exposed to emotion recognition or biometric categorisation systems must be informed about their use.
For most standard content marketing activities, this situation will be less common.
However, it may become relevant in areas such as events, market research, audience reaction analysis, digital experiences, or tools claiming to infer a person’s emotions from their voice, facial expression, or behaviour.
In addition, certain uses of emotion recognition or biometric categorisation are separately restricted or prohibited depending on the context.
For this reason, such a tool should not be implemented solely on the basis of the provider’s commercial claims.
The fact that a piece of content was created with the help of an AI tool does not automatically mean that it must display an “AI-generated” label.
Whether labelling is required depends on:
Let us consider a few examples.
A clearly stylised image used as an illustration for a blog article would not normally be mistaken for documentary photography.
In this case, the mere fact that it was produced using AI does not automatically create a visible deepfake-labelling obligation.
A photorealistic image showing a real CEO in a situation in which they were never photographed may mislead the public.
The material should be assessed as a potential deepfake and labelled when it may be perceived as authentic.
Before publication, the company should also consider other issues, including image rights, consent, and the reputation of the person depicted.
This is a clear example of content that may be mistaken for reality.
Using it without a visible disclosure may create serious risks related to manipulation and impersonation.
However, labelling does not automatically make every use acceptable. Issues concerning consent, misleading advertising, copyright, or reputational harm may still arise.
When AI is used as a working tool and the article is researched, verified, edited, and published under a company’s responsibility, this does not automatically result in a labelling obligation under the rule covering public-interest text.
However, the company must have an appropriate editorial process, rather than merely obtaining a formal approval immediately before publication.
A text about legislation that is generated and published automatically without human review may fall within the category of content intended to inform the public about a matter of public interest.
In this situation, the transparency obligation becomes relevant, while the risks go beyond labelling alone: incorrect legal information may lead readers to make poor decisions.
The disclosure must be clear, visible, and presented no later than the first relevant exposure or interaction.
Depending on the material, companies may use wording such as:
The European Commission has also published a set of icons that creators, publishers, and other organisations may use to label AI-generated content.
Using these icons is optional. However, the transparency obligation remains applicable whenever the conditions established by the regulation are met.
In other words, a company may choose the appropriate visual format, but it cannot turn a mandatory disclosure into a detail that is difficult to notice.
In June 2026, the European Commission published the final version of the Code of Practice on the marking and labelling of AI-generated content.
The Code includes measures for:
Adherence to the Code is voluntary.
Organisations that sign it may use the measures it contains to demonstrate compliance with the relevant AI Act obligations.
Companies that do not adhere may use other appropriate methods, but they must still be able to demonstrate that they comply with the rules.
For a marketing team, the Code can serve as a practical reference when establishing internal rules for creating, approving, and publishing content.
Compliance starts with understanding how your company uses AI.
Begin by identifying where AI is used in marketing activities:
Creating a list of tools is not enough. The purpose of each use must also be documented.
The same tool may produce a low-risk internal summary or a realistic video intended for a public campaign.
Risk is therefore determined not only by the technology, but also by the context in which it is used.
For published content, the process should include:
Human involvement should mean more than having someone click the publish button.
The team needs clear rules that can be applied to situations that may arise at any time.
For example:
For projects with a higher level of risk, retain information about:
Documentation supports both process consistency and the company’s ability to demonstrate accountability.
The AI Act also includes obligations relating to AI literacy, which have applied since February 2025.
Providers and deployers must take measures to ensure that people operating or using AI systems on their behalf have an appropriate level of knowledge and competence, taking into account their experience, the context, and the type of use.
Useful training should help the team recognise:
The AI Act is only one part of the broader compliance framework.
When a company uses AI tools, it must also consider data protection, information security, confidentiality, intellectual property rights, and contractual obligations.
In the absence of an approved environment and clear internal rules, employees should not enter the following information into public AI tools:
A simple rule can prevent many problems:
Do not enter information into an AI tool that you would not send to an external provider without first checking the contract, permissions, and security measures.
The exact policy should be adapted to the tools being used, account settings, supplier agreements, and the categories of data being processed.
The AI Act establishes several levels of penalties depending on the type of infringement, its severity, and the organisation involved.
For breaches of certain obligations under the regulation, fines may reach up to EUR 15 million or, for companies, up to 3% of the total worldwide annual turnover for the preceding financial year, with the applicable amount determined according to the rules of the regulation.
Different thresholds apply to other categories of infringement.
However, fines are not the only risk.
Content that misleads the public may affect:
For this reason, the responsible use of AI should not be treated as a simple legal box-ticking exercise.
Before publishing content created or modified using AI, check the following:
When the answer is unclear, the material should be reviewed before publication.
The AI Act does not prohibit the use of AI in marketing, nor does it require every piece of content produced with the help of a generative tool to be labelled automatically.
Instead, it requires greater clarity in situations where people may be misled or may not realise that they are interacting with an artificial system.
For companies, this is also an opportunity to build better processes:
Transparency and accountability do not reduce the value of AI. They create the conditions in which the technology can be used with greater confidence.
At Beans United, we understand the value of human contribution and make sure that every piece of content, analysis, research project, or campaign is supported by a rigorous research process and authentic creative work.
At the same time, we embrace technology and use it to provide our clients with innovative solutions that help them increase visibility, attract more qualified leads, streamline internal processes, and achieve their business goals.
In our work, we aim to ensure that the initiatives, materials, and AI systems we use comply with the applicable requirements of the AI Act and with the principles of responsible artificial intelligence.
Not automatically. The answer depends on the purpose of the article, the topic, the extent of AI involvement, whether human review took place, and whether editorial responsibility was assumed. For texts concerning matters of public interest, an exception applies when the material has undergone human or editorial review and a natural or legal person assumes responsibility for its publication.
Not every image automatically requires a visible label. The obligation is particularly relevant when the material may be mistaken for reality, such as when it realistically depicts a person or an event that never occurred.
In general, users must be informed that they are interacting with an AI system. The exception covering situations in which this is obvious should be interpreted narrowly. Explicit disclosure is therefore usually the safest approach.
Yes. The AI Act does not prohibit the use of generative tools for research, ideas, text, images, or automation. The company must assess the use case, verify the outputs, and comply with the applicable transparency obligations.
Not in every case. The label must be clear, visible, and appropriate for the type of material. In addition, labelling does not replace fact-checking and does not resolve other legal issues, such as a lack of consent or the unauthorised use of data.
Responsibility depends on the roles and actions of the parties involved. Providers and deployers have different obligations, while the company publishing the material may retain editorial, contractual, and legal responsibilities even when it uses a tool supplied by another organisation.
The EU icons are optional. When a transparency obligation applies, the company must comply with the requirement, but it may use other appropriate methods of disclosure.
No. Under the revised European timeline, the rules for certain high-risk systems, including those used in recruitment, education, biometrics, or critical infrastructure, apply from 2 December 2027. For AI systems integrated into certain regulated products, the deadline is 2 August 2028.