The AI labeling requirement is here: What takes effect in August 2026 and how you can evaluate AI-generated images

This article summarizes the current guidelines for businesses and is not a substitute for legal advice.

 

Just two years ago, many AI-generated works gave themselves away: images of people with six fingers alongside texts filled with fabricated nonsense that read as if the author had relinquished both his or her critical judgment and the pursuit of new perspectives.

By now, everyone has likely noticed that the quality of AI-generated content has reached a level where it is difficult to distinguish it from human-created content. This makes it all the more important to restore some transparency and, by labeling AI content, allow consumers to decide for themselves how much of it they wish to consume.

The good news: This labeling requirement has been in effect throughout the EU since August 2 as part of the EU AI Act. It is intended to curb deception and misinformation and foster public trust. However, the labeling requirement brings new uncertainties, because the bad news is: The regulation lacks clarity. In some cases, it remains a matter of interpretation as to what requires labeling and what does not. There’s only one solution: sharpen your own judgment.

For companies, labeling AI-generated content has not only a legal dimension but also a communicative one: Those who are open about their use of AI strengthen the trust of their customers and the public. In the future, case law will provide further guidance through landmark rulings on how to comply with these requirements in practice in a legally sound manner. Until then, the rule of thumb is: When in doubt, label the content and view this transparency as an integral part of your company’s communications.

AI-Generated Texts—The Big Exception

As a communications and PR agency, we are naturally particularly interested in how the labeling requirement affects texts. And this is where the major exception comes in. An AI-generated or AI-manipulated text must be labeled only if it informs “the public about matters of public interest” and if that text has not been subject to human review or editorial oversight.

Conversely, this means that as soon as an editorial team takes responsibility for the text, the requirement to label it as AI-generated no longer applies. Standard product and marketing texts, press releases, or specialized articles that have been reviewed by humans are therefore not affected. Publishing an AI-generated text without review is, in any case, a no-go according to our quality standards.

We primarily use AI as a sounding board; we write most of the content for our clients ourselves and incorporate AI-generated additions only where it makes sense. People are better at striking the right tone than machines. After all, effective public relations relies on context, connections, industry knowledge, and an understanding of the target audience. No AI can replace these strengths.

Deepfakes Must Generally Be Labeled

Visual and audio content is subject to stricter requirements than text: Deepfakes must always be labeled as such. These are AI-generated or manipulated images, videos, or audio clips that resemble real people, places, objects, or events and could be mistaken for the real thing.

A prominent example: the image of the pope in a down jacket that went viral in early 2023. So when PR agencies or media outlets use realistic-looking AI images—including AI-generated stock images—to illustrate articles, these images must be labeled, as readers might mistakenly believe them to be real.

Since uncertainties often arise, particularly when evaluating images, the following examples from the B2B world are intended to help you get a sense of the criteria that must be used to label images.

Example 1: Real Product in an AI-Generated Environment

A pump manufacturer photographed its new product at its own plant after final assembly: The pump is standing on a pallet in front of a silver warehouse wall. For the campaign, the marketing team used AI to generate a suitable setting. In the final image, the pump is shown installed in a chemical plant. Everything fits together so well that there’s no indication it was a staged shot.

Required to be labeled? Yes

Reasoning: The image suggests that the photo was actually taken at the location shown. However, the installation scenario is fictional. There is a high risk of misleading viewers. References play a major role, especially in the B2B sector, so credibility is a top priority.

Example 2: Sketch / Engineering Drawing

A plant engineering company is promoting a technical article on LinkedIn about the modernization of production lines. The cover image is an AI-generated, blueprint-style illustration in which scribbled lines suggest a manufacturing facility with piping. The image is intended to help users more easily associate the post with the topics of planning and engineering as they scroll through their feed.

Required to be labeled? No

Reasoning: Because they are sketches, such images cannot be mistakenly classified as authentic or truthful.

Example 3: Classic Image Editing

An IT security provider uses a photo of its booth at a trade show for a blog post. Because the lighting was poor, the image is edited, in part using AI-powered tools. Noise is removed, exposure and white balance are corrected, and the frame is cropped to a tighter composition.

Required to be labeled? No

Reason: No significant change to the image's message resulting from optimizations such as noise reduction, exposure compensation, cropping, or adjustments to sharpness.

Example 4: Drawing / Comic / Cartoon

Take a look at the cover image for this post. It was generated by AI and includes a voluntary label to draw attention to the topic in the blog overview. However, it is not required to be labeled as such.

Required to be labeled? No

Reasoning: Viewers can tell from the drawing style that this is fiction. The risk of deception is very low.

FAQ – Frequently Asked Questions About the AI Labeling Requirement:

Who regulates the AI labeling requirement?

The labeling requirement is set forth in Article 50 of the AI Act, an EU regulation.

Who is affected?

All organizations, regardless of their size, that use AI in a way that allows its results to reach people in the EU.

Since when has the labeling requirement been in effect?

The AI labeling requirement has been in effect since August 2, 2026. A transition period through December 2, 2026, applies to technical markings, such as watermarks, added by providers.

What needs to be labeled?

Chatbots; deepfakes (video, images, and audio); AI-generated texts of public interest that have not been editorially reviewed; AI that recognizes emotions or analyzes biometric features.

What does not need to be labeled?

Texts that have been reviewed by humans or edited, as well as images that have no connection to reality. These are images that are obviously not realistic depictions and therefore cannot be mistaken for real (a matter of interpretation).

How should AI-generated content be labeled?

Through labels or icons. The European Commission’s website provides suggested icons here. These are not mandatory; you may use your own labels. Important: The label must be visible upon first contact with the medium; for audio, this means a spoken notice at the beginning. Website text should ideally be marked with a prominent box above the text. A notice in the Terms and Conditions or in the footer is not sufficient. For images, the label in the caption or source citation is sufficient. In videos, the icon must be clearly visible on screen.

What are the potential penalties?

Authorities can impose fines of up to 15 million euros or 3 percent of global annual revenue for violations.

Do I have to retroactively tag my posts?

No. Anything created and published using AI before August 2 does not need to be retroactively labeled.

 

Sources:
EU Icons for Labeling AI-Generated Content | Shaping Europe's Digital Future

Quick Facts: Transparency Rules for AI Systems | Shaping Europe's Digital Future

AI Labeling Requirement in 2026: The Complete Guide (Art. 50 of the EU AI Act) | AIActify

 

Caption/Photo Credits: Cover image generated by AI