INSIGHTS
What the EU AI Act means for brands, agencies & creators
The hardest part of the new AI transparency rules is not the obviously fake image or the chatbot that clearly says it is AI. It is the everyday gray zone: a campaign visual generated by a production partner, a creator asset polished with AI, a synthetic product scene that looks real, or a social post that passes through several tools before it goes live. For brands, agencies and creators, the question is no longer only what can we make with AI? It is when does the audience need to know and how do we make that clear?
Since 2 August 2026, the EU AI Act has made AI transparency part of everyday brand and creator content. The rule is not that every use of AI needs a label. A brainstorm, caption draft or small edit will not automatically trigger disclosure. What matters is whether the audience can clearly understand what is real, what is AI-made, and whether they are interacting with a person or a system. That is why this guide looks at two simple questions: when do you need to disclose AI, and how do you make that clear in the formats people actually see?
WHEN TO DISCLOSE AI? / HOW TO DISCLOSE AI?
WHEN TO DISCLOSE AI?
1. DEEPFAKES
For social and campaign content, deepfakes are one of the clearest disclosure risks. Under the AI Act, this means AI-generated or AI-manipulated image, audio or video content that could falsely appear authentic or truthful. That can include a synthetic product scene, a realistic AI person, a changed face or voice, or a visual of a moment that never happened.
Not every AI edit needs a label. Small technical adjustments such as resizing, compression, limited color correction or privacy blurring will usually fall outside the duty if they do not change the substance or meaning of the content.
The threshold changes when AI alters what the asset communicates: removing or adding people or objects, changing a face or voice, creating a realistic scene that did not happen, or making a product look better than it actually is.
→ What to check: ask whether the AI edit changes what the audience believes they are seeing. If it does, the label likely belongs on the asset, not only in the caption.
WHEN TO DISCLOSE AI?
2. AI-WRITTEN TEXT
Not every AI-assisted caption, product description or campaign line needs an AI label.
For written content, the key question is what the text is used for. The AI Act focuses on AI-generated or AI-edited text that informs the public about matters of public interest, such as politics, public health, safety, the environment or public administration.
Most brand copy does something different. A product description, campaign caption or social post that promotes a product or brand usually falls outside this specific text duty.
Some brands should be more careful. Healthcare, finance, insurance, public services or brands communicating around social issues can move closer to public-interest territory, depending on the message.
There is also a role for human review. If AI helps draft public-interest text, a real person or organization can take editorial responsibility, but that review has to be meaningful. A quick spelling check or automatic approval is not enough.
→ What to check: ask what the text is doing. Is it brand communication, or is it informing the public about a societal or sensitive topic? If it is the second, make sure human review comes last, or clearly disclose the use of AI.
WHEN TO DISCLOSE AI?
3. CHATBOTS
When a brand uses a chatbot, voice bot, AI agent or avatar, users need to know they are interacting with AI. That disclosure should be clear from the start of the interaction, not hidden in terms and conditions.
This matters for customer service, campaign environments, websites, event tools or any branded experience where AI speaks directly to someone. A bot can have a strong brand voice, but it should not create confusion about whether a human is on the other end.
If the conversation is long, sensitive or aimed at younger or more vulnerable users, one opening line may not be enough. In those cases, the AI disclosure should be repeated where it makes sense.
→ What to check: make sure users immediately understand they are interacting with AI, especially when the bot looks, sounds or behaves in a very human way.
WHEN TO DISCLOSE AI?
4. SYSTEMS THAT READ PEOPLE
Not every AI transparency question appears in the final post. Some appear earlier, in the tools brands use to research, test or analyze audiences.
This matters when an AI system tries to infer someone’s emotional state or categorize people based on biometric traits. In those cases, the people exposed to the system need to be informed, and the data must also be handled in accordance with the GDPR.
For most brand and creator content, this will not be a daily publishing issue. But it can become relevant in research, creative testing, audience analysis, retail experiences, event technology or interactive installations.
→ What to check: does the tool only help create content, or does it analyze something about the person using it? If the tool reads people, people need to know.
HOW TO DISCLOSE AI?
5. A CAPTION IS NOT ALWAYS A LABEL
A caption can help explain AI use, but it should not always carry the full disclosure. On social, the first impression often happens before anyone reads the caption. Captions can be collapsed, hashtags can be missed, and platform AI badges do not automatically replace the brand’s own disclosure responsibility.
That matters most when an image, video or voice could be mistaken for something real. In that case, the disclosure should not only sit below the post. It should be visible on the asset itself, at the moment the audience sees or hears it.
The point is not to over-label every AI-assisted asset. It is to make sure that when AI changes how something may be understood, the clarification is not hidden in a place people might never open.
→ What to check: ask whether the AI-made or AI-manipulated element is clear without reading the caption. If not, the disclosure needs to move closer to the content itself.
HOW TO DISCLOSE AI?
6. EVERY FORMAT, ITS OWN PLACEMENT
A feed post, story, reel, video and carousel are experienced differently, so the label should follow the format.
Feed post: if the image or video could be mistaken for something real, place the label directly on the visual. Add a short disclosure in the caption, ideally above the hashtag block.
Story: because there is no separate caption field, use a text overlay on the story itself.
Reel or video: show the label at the start of the video, before the AI-made or AI-manipulated element appears. Add more detail in the description when needed.
Carousel: treat every slide separately. If a slide contains AI-made or AI-manipulated content, that slide should carry the label. The full disclosure can appear once in the caption.
For AI use without deepfake risk, a caption mention can be useful as a house practice, but it is not automatically required. First determine whether the content actually needs a disclosure.
→ What to check: review the content in the format where it will appear. A label that looks clear in an approval deck may be easy to miss once the post becomes a story, reel or carousel.
AI TRANSPARENCY SHOULD MOVE EARLIER
The biggest shift is not that every AI-assisted asset suddenly needs a label. It is that AI transparency can no longer be decided only at the moment of publishing.
If content could be mistaken for something real, the disclosure should be considered from the start: in the brief, the creator guidelines, the production process and the approval flow. That matters even more when agencies, creators, freelancers or production partners are involved, because outsourcing the work does not automatically outsource the responsibility.
OUR VIEW
Transparency protects trust
AI will keep making content faster, easier and more scalable. But in creator marketing, speed is not the only value. Trust is.
The goal is not to label everything out of fear. The goal is to understand when clarity matters. If AI changes what an audience thinks they are seeing, transparency becomes part of the relationship.
That is why AI disclosure is not just a compliance question. It is a trust question.
And trust is still what makes creator marketing work.