Between Legal Obligation and Voluntary Transparency: The AI Labeling Requirement
ChatGPT, Claude, Perplexity, Copilot, and Gemini are just a few examples of AI tools that have become an integral part of everyday life. Everyone knows them, and for many, using AI in daily life has become second nature.
According to a recent study by the digital association Bitkom, one in three people already uses AI at least once a week, and 15% of respondents even said they consult artificial intelligence daily. The use of AI has gained particular traction in the workplace. Just under half of the working population uses AI while at work.
Would you like to take a look at this study yourself? Then feel free to check it out here!
The Rapid Development of AI Tools
The intensive use of AI ensures that these tools continue to evolve at a breakneck pace, not least because of the immense amount of data they are fed every day. The tools receive continuous updates and are becoming increasingly powerful. Let’s take a look at ChatGPT, for example. Although it is essentially a free tool, even there, image generation is now incredibly powerful. And this is in an application that’s designed primarily for text-based support. Specialized AI software for generating images and videos is even more powerful by a wide margin. And to be honest: Even for a “digital native” like me, it’s becoming increasingly difficult at times to distinguish AI-generated content from human-created media. AI is making fewer and fewer errors in images and videos. Whereas before a person in an image might have had six fingers instead of five, or the background might have looked like a jumble of pixels, such errors are becoming increasingly rare as a way to distinguish between reality and AI.
You might be asking yourself the same question I am: How, then, can we still tell the difference between generated content and genuine, authentic content? The better AI becomes and the harder it is to distinguish between artificial and human intelligence, the more important it becomes to label generated content as such. And that’s exactly where the European Commission is stepping in with the so-called “AI Act.” It went into effect on August 1 of this year.
The European Commission's labeling requirement
Essentially, this new legislation requires the labeling of AI-generated content in the context of public information provision, including AI chatbots and other AI systems. AI can be used, for example, for emotion recognition or biometric categorization. This might be the case, for instance, in call centers, where a voice is analyzed using AI to infer the emotions of the customer making the call from their voice. The requirement to label AI-generated content therefore applies in particular in the public interest and when there is a high risk of confusion with real people.
Consequently, this labeling is especially relevant for image, video, and audio media. In this context, the misuse of AI can lead to the creation of deepfakes. This applies not only to fake videos of politicians but also to generated advertising media or other content where there is a risk of confusion with existing people, places, objects, institutions, or events. It does not necessarily have to involve real people or similar entities; it is sufficient if the generated content looks so authentic that a user perceives it as real and is thereby misled.
Regulations regarding text generation are somewhat less strict in this regard. Here, the European Commission clearly states the limitation that, for now, only texts published without prior human review and editorial revision are subject to this labeling requirement. And, most importantly, this labeling requirement initially applies only to institutions serving the public interest, such as those in the healthcare sector or providing legal content.
Why Labeling Makes Sense Even Without a Mandate
Even for a company that uses AI to create marketing and informational materials about its own products and services, an AI disclosure can certainly be useful. I don’t know about you, but even though today’s AI tools are getting more and more powerful, I still feel like I can tell when a text was written by artificial intelligence. And of course, especially in marketing, a lot of time can be saved when creating text with the help of AI. “Garbage in—garbage out.” If you provide good prompts, you’ll usually get good results. But presenting AI-generated text as if it were handwritten leaves a bad taste in my mouth. Which brings us to the reason why labeling content can be beneficial even for companies not directly involved: transparency.
The fact that texts are structured, written, or edited using AI is already almost standard practice, at least in my circle. AI-generated texts aren’t bad in and of themselves. But, in my opinion, it all comes down to how they’re presented. Essentially, AI generates texts based on existing data it draws from specifically inputted materials or publicly available sources on the internet. So it seems only fair to label a text when it wasn’t written by the company itself. This creates transparency. Reading a text that is obviously written by AI based on its writing style, but then seeing a photo of a company employee at the end of the text who is credited with the article, just doesn’t add up. A transparent explanation of how a text was created fosters trust and demonstrates authenticity.
How do we handle this at PROTAKT?
Here at PROTAKT, we neither offer our own AI bots nor do we plan to publish deepfakes. And yet, we, too, use AI. Our experts are developers and business analysts with a wealth of knowledge and an even busier schedule. That’s where ChatGPT, Copilot, and the like can often help organize existing expertise into the right textual structure in a time-saving way. Our #PROTAKTVoices on LinkedIn also use these little helpers to optimize and create their texts or graphics. But all our texts have one thing in common: They are not published without human review and, if necessary, editorial revision. For our blog, we have an editorial team (of which I am a member), and for our Voices, we have established internal criteria for disclosing AI usage. Yes, we are not subject to any potential sanctions. Nevertheless, we want to be transparent with you, our readers, about how and by whom our texts were created. Accordingly, we’ve decided to classify our texts and graphics into three categories: “Generated,” “Modified,” and “No AI.”
For all three categories, we’ve developed corresponding labels that allow you to classify the texts directly.
The “Generated” label indicates whether a text or graphic was created entirely by artificial intelligence and uploaded without further review or editing.
For us, “modified” means that AI was used to help draft and create the text or graphic, but that the authors conducted a thorough review and, where necessary, our team made revisions.
The “No AI” category lives up to its name: The content here was written and edited in-house.
However, when it comes to our technical expertise and competence, the type of text we produce is irrelevant. You can rest assured that the know-how and expertise we provide in our content always come directly from our team. We use artificial intelligence to support our processes, not as a replacement for them. AI has no experience. It understands rational and structured processes, can suggest solutions based on an existing database, and can follow instructions. But it hasn’t been involved for years in a client project where things don’t always go according to plan—where requirements, structures, and circumstances change. It knows only raw information; it may be able to link it to other information, but it doesn’t know that the most rational approach isn’t always the best one. Our colleagues know this. AI cannot replace human experience, complex thought processes, or—necessarily—flexible new approaches to problem-solving.
Our conclusion
The rise of AI is transforming our processes, the way we work, and our efficiency. But it also highlights just how important human-created content, expertise, and transparency are—perhaps more than ever before. With the right tools, any information on the internet can be accessed and made available at the click of a button. “New” content can be created from existing content. It has never been faster to generate an image, a video, or a text without having to invest a significant amount of time and effort.
However, the AI Act aims to prevent this work from being sold as one’s own. A fake image or a text containing misinformation can have far-reaching consequences. Opinions can be formed based on false data. And if there’s no person behind it who is accountable for this content, there’s no one who can be held responsible. The AI labeling requirement is therefore intended to create transparency and counter the spread of misinformation or the misuse of AI. How would you feel if a video of you suddenly started circulating online that you didn’t record yourself, but was instead created by AI? The AI Act is designed to counter exactly such actions, particularly with regard to opinion-shaping content and deepfakes.
But even without a legal labeling requirement, the question remains: how responsibly is AI being used? Even if a company like PROTAKT is not subject to the labeling requirement, we still want to be transparent with you. It’s important to us that you can understand where the information you’re currently reading or viewing actually comes from. By presenting our content transparently, we want to make it clear that we take a conscious approach to artificial intelligence. We see it for what it is: a support tool in our busy workday and a means of increasing efficiency. It is not a substitute for human experience or expertise. Behind every text, every image, every graphic, and every audio track, there is ultimately still a PROTAKT team member or team. Whether it’s a website, blog, or social media post.
Stay up to date and subscribe to our PROTAKT blog!
About the author
Katharina Falkenstein · MARKETING
Having always been passionate about language, communication, and creative ideas, I’ve found my calling in marketing. Since September 2023, I have been supporting the PROTAKT marketing team as a dual-track student while gaining valuable experience through my business administration studies.
Here at PROTAKT, I work closely with our solutions and services, manage our blog, and keep an eye on website and social media topics. In doing so, I learn something new every day and enjoy sharing this knowledge through clear and practical posts.