Artificial intelligence is being used in marketing more and more often. Companies use it to create images, videos, advertising visuals, voiceovers, product demos and even entire campaigns. This raises an important question: when do you need to disclose that content was created or edited using AI?
The European AI Act introduces transparency rules aimed at helping people recognize when they are viewing artificially created or significantly altered content. This does not mean that every post created with the help of an AI tool needs a big warning label. However, realistic AI content that may appear to be real should be labelled clearly, visibly and understandably.
When the rules will apply
The transparency obligations for AI content under the AI Act will generally start applying from 2 August 2026. However, this does not mean that labelling does not need to be addressed before then.
Social platforms such as Meta, TikTok and YouTube already have their own rules. They may require realistic synthetic content to be labelled, add their own label, limit the reach of a post, reject an ad or remove the content.
AI content may also be subject to other rules, such as consumer protection, the prohibition of misleading advertising, personality rights, copyright or rules on political communication. That is why it makes sense to set up a simple labelling system now.
Who AI content labelling applies to
The AI Act does not apply only to technology companies. Obligations may also affect brands, agencies, influencers, e-shops, media outlets or content creators who use AI outputs and publish them to audiences in the EU.
Typical examples include companies using AI visuals in advertising, agencies creating AI videos for clients, e-shops working with AI product images or brands publishing reels that combine real and AI-generated footage.
Importantly, the AI Act is not tied to a specific platform. It does not say that you must label content on Instagram but not on LinkedIn. The rule is general: if you publish a certain type of AI content for an audience in the EU, people must be able to understand that the content was created or edited using artificial intelligence.
When AI content needs to be labelled
The basic rule is simple: if AI content may appear to be a real recording, a real person, an authentic event or a real result, it should be labelled.
Labelling is especially appropriate for realistic images and videos that an average user could mistake for reality. Deepfakes and synthetic voices require particular attention. If AI imitates the appearance, voice or behaviour of a specific person, labelling is very important. At the same time, the consent of the person concerned and the protection of their personality rights must also be addressed.
The risk is usually lower for content that is clearly stylized, abstract, fairytale-like or obviously unrealistic. Even there, however, labelling may be appropriate, especially in corporate communication, where transparency is an advantage.
AI texts and publicly important topics
The AI Act also addresses certain text outputs. If AI generates or significantly edits text intended to inform the public about matters of public interest, transparency may be required.
This may include topics such as politics, elections, health, finance, safety or public events. Exceptions may apply where the text has undergone human review, is subject to editorial responsibility and someone clearly takes responsibility for it.
For ordinary marketing texts, such as product descriptions, advertising slogans or social media posts, it will usually not be necessary to label every use of AI assistance. However, it always depends on the context and whether the content could mislead users.
On which platforms AI content should be labelled
The rules may apply to all common online channels, such as Instagram, Facebook, TikTok, YouTube, LinkedIn, X, websites, blogs, newsletters or online advertising. The main difference is that some platforms have their own tools for labelling AI content. Meta uses labels such as “AI info”, TikTok has a label for AI-generated content and YouTube requires disclosure of realistic synthetic or altered content.
If the platform offers such a label, it is advisable to use it. However, the platform label alone may not always be sufficient, especially if it is not very visible or if the content is risky.
Is a platform label and a sentence in the caption enough?
For most ordinary corporate and marketing content, a good standard is a combination of two steps:
- use the platform’s AI label, if available,
- add a clear sentence to the caption.
This type of labelling is simple, understandable and usable across platforms. At the same time, it does not unnecessarily disrupt the creative format of the post.
It is important that the label is not hidden deep in the text where nobody will see it. Ideally, it should be visible in the caption without requiring users to expand the text, or at least placed as close to the beginning as possible.
Does the label have to be directly in the video?
Not always. If you use the platform label and add a clear sentence in the caption, this will often be enough for ordinary content.
Labelling directly in the video is appropriate mainly when the AI footage could easily be mistaken for reality. For example, if you show AI footage of an event that looks real, an AI customer, an AI employee, a realistic demonstration of a service result or a synthetic voice of a specific person.
In such cases, it is safer to add a short text directly into the video. It does not have to be a large, disruptive notice across the entire screen. A discreet but readable label next to the relevant scene is enough.
What if reels combine real and AI footage
Reels, shorts and other short videos often combine real filmed material with AI illustrative footage. This can be problematic because viewers may not recognize which parts are real and which were created using AI.
The most practical solution is to label the entire video as containing AI. If the AI footage is only supplementary and does not risk materially changing the meaning of the message, it is usually not necessary to label every individual shot directly in the video.
However, if an AI scene looks like an authentic recording of reality, it is better to label it directly in the image as well. For example, if a video shows a real event but includes an AI-generated crowd or an AI scene from the venue, viewers could get a false impression of the reality of the event.
What penalties may apply for not labelling AI content
Violations of the European Artificial Intelligence Regulation, the AI Act, may lead to fines divided into three main categories:
- Up to EUR 35 million, approximately CZK 880 million, or up to 7% of total annual worldwide turnover, whichever is higher, for violations of prohibitions relating to banned AI practices, such as social scoring systems or manipulative techniques.
- Up to EUR 15 million, approximately CZK 375 million, or up to 3% of total annual worldwide turnover, for failure to comply with other key obligations, such as obligations relating to high-risk AI systems, data governance, technical documentation or human oversight.
- Up to EUR 7.5 million, approximately CZK 187 million, or up to 1% of total annual worldwide turnover, for providing false, incomplete or misleading information to supervisory authorities.
In practice, however, not every missed label will automatically result in the maximum fine. The severity, scale, intent, impact on the public and whether it was a one-off mistake or a systematic practice will all matter.
In addition to the AI Act, other consequences may also arise: removal of the post, limited reach, rejection of an ad, account warnings, issues with misleading advertising or intervention due to personality rights.
The greatest risk is content that can genuinely mislead people. Examples include fake testimonials, deepfakes of real people, AI demonstrations of service results or visuals that create a false impression of a product, company or event.
Practical recommendations for companies
For corporate communication, it is best to set up simple internal rules:
- Always label realistic AI images, videos or voices.
- For a mix of real and AI footage, label the entire video as containing AI.
- Use the platform label if it is available.
- Add a clear sentence to the caption.
- For risky scenes, add a label directly into the video.
For ordinary reels, an AI platform label and a sentence in the caption will therefore usually be enough. For more sensitive or more realistic content, it is better to also add text directly next to the AI footage.
Conclusion
Labelling AI content under the AI Act does not have to be complicated. Just ask: could viewers think they are seeing a real recording, a real person, a real event or an authentic result?
If yes, label the AI content.
For most corporate content on social media, a good minimum standard is to use the platform’s AI label and add a clear sentence in the caption. For reels combining real and AI footage, it is appropriate to state that the video contains a combination of real and AI-generated footage. Labelling directly in the video should mainly be used for parts that may appear real and could mislead viewers.
A transparent approach helps reduce legal risk, comply with platform rules and strengthen audience trust.