Labeling AI Content: The Deadline Is Ticking

On July 22, 2026, at 6:00 p.m. CEST, a window closes. Anyone who hasn't signed by then won't appear on the first official list of early signatories of the Code of Practice on Transparency of AI-Generated Content - the voluntary commitment with which the EU Commission flanks the statutory labeling obligations under Article 50 of the AI Act. And while this date is currently making the rounds in professional communities, a second, far more important clock is ticking almost unnoticed: from August 2, 2026, the transparency obligations themselves apply - binding, with no grace period, and unchanged by the much-discussed Digital/AI Omnibus package.

For marketing leaders in mid-sized companies who have long been using AI agents to produce blog articles, social media posts, and ad copy, this raises a legitimate question: does every Instagram post now need a label? The short answer is: no, but it depends very precisely on what you publish and how. This article breaks down what the new Transparency Code actually requires, where the statutory obligation begins, where it ends - and what you should concretely do in the coming weeks.

What the Transparency Code is - and what it isn't

First, a clarification of terms, because a lot gets mixed up here in practice: there are two entirely different "Codes of Practice" in the European AI regulatory landscape. One is the Code of Practice for General-Purpose AI Models, signed by major model providers such as Anthropic, Google, and OpenAI (Meta declined) - it concerns developers of foundation AI models. The other, which this article is about, is the Code of Practice on Transparency of AI-Generated Content. It targets anyone who provides or publishes AI-generated content - potentially including your company, as soon as you use chatbots, AI-generated images, or AI-assisted text.

After two draft rounds, the EU Commission published the final Code in June 2026. Unlike the AI Act itself, it is formally voluntary. But it directly addresses the statutory obligations under Article 50(2) and (4) of the AI Act, and signatories who implement its measures benefit from what's called a "Presumption of Conformity." That means: supervisory authorities focus primarily on compliance with the Code for signatories, instead of examining each individual case separately. For companies that produce AI content strategically and visibly, this is a tangible advantage in legal certainty and administrative burden.

The three core obligations - short and concrete

The Code translates the statutory requirements from Article 50 of the AI Act into three practical obligations:

1. Disclosure of AI interaction: Users must be told that they are communicating with a chatbot, virtual assistant, or AI agent - unless this is obvious anyway to a "reasonably informed" person. If you run a customer-service chatbot on your website, a clear notice is required.

2. Labeling of deepfakes: AI-generated or manipulated images, audio, and video content must be identifiable as such. This applies to advertising videos with synthetic voices or AI-generated faces, as well as imagery created by an image generator.

3. Labeling of text on matters of public interest: AI-generated or manipulated text published to inform the public on matters of public interest must be labeled - unless it has undergone human editorial review before publication and an identifiable person or organization takes editorial responsibility for it.

For practical implementation, the Code proposes a standardized, high-contrast EU "AI" icon, optionally supplemented by a short additional text such as "Created with AI." Placement should be clearly visible - for example in the headline, above the text, or in the legal notice - and for social media thumbnails directly as an overlay on the image, so the label is preserved when shared. If only a section of text is AI-generated, labeling that section is sufficient.

Does this really affect you? The important distinction for marketing

This is where a lot of unnecessary panic has spread among mid-sized companies in recent weeks. The labeling obligation for text under Article 50(4) does not automatically apply to every AI-assisted blog article, product description, or ad copy. According to the prevailing interpretation of legal experts, it applies exclusively when two conditions are met simultaneously: the text addresses a matter of public interest, and it was published without human editorial review.

For the vast majority of marketing texts, product descriptions, and advertising content that a mid-sized company publishes, neither condition typically applies - especially not if a human reads, edits, and approves the text before publication, which should be standard in professional content workflows anyway.

The statutory labeling obligation for text targets opinion-forming and public information without editorial control - not every AI-assisted piece of ad copy. Anyone who reviews their content editorially before publication generally falls outside the statutory labeling obligation for text.

Where mid-sized companies genuinely need to act can be clearly named: first, with their own chatbots or virtual assistants on the website, because the disclosure obligation toward users applies here. Second, with content that has genuine news or opinion-forming character and is published without human editorial oversight - for example, an automatically generated "news" feed with no review step. And third, with any deepfake-like image, video, or audio material, regardless of topic.

Pure advertising copy and product descriptions generally do not fall under the labeling obligation according to this interpretation. That takes a lot of pressure out of the debate - but it doesn't remove the need to go through your own content mix cleanly once and check where editorial-like or opinion-forming content is actually being created.

An example for context: a blog article by your company about the benefits of your own software solution is advertising - even if it was created with AI assistance and contains facts, figures, or market observations. A post that presents itself as a neutral assessment of a political or societal topic, without an identifiable person taking editorial responsibility for it, on the other hand, falls into a different category. It's exactly at this boundary that a second look at your own content strategy is worthwhile: anyone who regularly publishes opinion pieces, market assessments, or commentary on regulatory topics - as this very article is - should document the editorial review step and clearly name a responsible person. That's good journalistic standard practice anyway, but the Transparency Code additionally turns it into a legal shield.

The deadlines at a glance - and what the Omnibus does not change

To avoid confusion, here are the relevant dates summarized compactly:

July 22, 2026, 6:00 p.m. CEST: Signup deadline for the first official list of early signatories of the Transparency Code. Registration is done via a signed form sent to a dedicated functional mailbox of the EU Commission and must be signed by an executive with power of representation. Anyone who misses this date can still sign later - but will not appear on the first list, which is published before August 2, 2026.

August 2, 2026: The statutory transparency obligations under Article 50 of the AI Act take effect. This date was not postponed by the Digital/AI Omnibus agreement of May 7, 2026 - unlike the obligations for high-risk AI systems under Annex III, which were postponed to December 2027 and August 2028 respectively. It's worth communicating this distinction clearly within your own company, because in public debate the postponed high-risk deadlines and the unchanged transparency deadline are often conflated.

December 2, 2026: For generative AI systems already on the market before August 2, 2026, an extended transition period until this date applies to the machine-readable labeling obligation under Article 50(2).

The Digital/AI Omnibus has indeed brought relief in other areas: simplified technical documentation, more proportionate fines, and less prescriptive requirements for quality management systems, previously reserved for small and medium-sized enterprises (SMEs), are set to be extended to a broader group of smaller mid-sized companies. For the transparency obligations themselves, however, this changes nothing about content or timing.

Specifically, the relief already available to small and medium-sized enterprises - simplified technical documentation, more proportionate fines, less prescriptive requirements for quality management systems - is set to apply in future to a broader group of smaller mid-sized companies as well (so-called Small Mid-Cap Companies, SMCs), which previously fell between the SME threshold and the full obligations of large corporations. For the transparency obligations under Article 50, this changes nothing about their fundamental applicability, but it can noticeably ease the practical implementation - for example, in the scope of required documentation.

Importantly, to clarify what actually threatens in case of non-compliance: fines are not imposed for a company not signing the voluntary Transparency Code. They are imposed for actual violations of the statutory obligation under Article 50 of the AI Act. The fine range is considerable: up to €15 million or up to 3 percent of the company's worldwide annual revenue, whichever is higher. A separate cap of up to €750,000 applies to EU institutions and authorities. National market surveillance authorities are responsible for enforcement.

Not signing the voluntary Code is thus, by itself, sanction-free. The presumption of conformity that the Code brings for signatories is an advantage - not an obligation. Anyone who cleanly fulfills the statutory obligations under Article 50 even without signing is also compliant. It's just that then every individual measure may need to be individually justified to the authority in case of doubt, instead of being able to rely on the recognized Code measures.

To sign or not? A sober weighing of the options

For many mid-sized companies, it's worth weighing both sides before the deadline on July 22 passes.

Arguments in favor: The presumption of conformity noticeably reduces administrative burden vis-à-vis supervisory authorities. A signature is also a visible trust signal to customers and partners - especially for companies that openly and prominently communicate AI content production as part of how they work. Anyone who, for example, regularly produces social media content with AI agents like Natascha or editorial content with other agents can position an early signature as part of a transparent, responsible AI strategy.

Arguments against: The Code is voluntary but ties you to concrete implementation measures - labeling formats, internal processes, documentation. For small teams, the associated documentation and process overhead can, in individual cases, outweigh the benefit of being on the early signatory list. And since signing later remains possible at any time, waiting doesn't cost you the ability to join the Code - only the PR bonus of being on the very first published list.

There's no blanket right answer here. Companies that produce AI content strategically and at scale tend to benefit more from an early signature than companies that use AI only selectively and as a supplement in marketing. In any case, it's important that an executive with power of representation signs the form - that requires internal lead time, and the time remaining until July 22 is limited.

How to implement labeling in practice

Regardless of whether you sign or not: anyone using AI in content production would do well to establish a clear, repeatable labeling workflow now - for cases where the obligation actually applies, and as a general trust signal beyond that.

Specifically, this means: determine which of your content formats could potentially fall under the text labeling obligation - typically editorial-like posts without a review step, not classic product advertising. For these formats, define a standard badge modeled on the proposed EU "AI" icon, supplemented by a short additional text such as "Created with AI," and place it visibly - in the headline, at the start of the post, or in the legal notice. For social media images, the label belongs as an overlay directly in the image, so it remains visible even when shared. And for every chatbot or virtual assistant on your website, ensure a clear, unambiguous notice at the start of the conversation.

A short checklist for the coming weeks summarizes the essentials: first, check by July 22 whether signing the Transparency Code makes sense for your company, and arrange in good time for signature by an executive with power of representation. Second, go through your content calendar and mark which formats have editorial-like character and which are pure advertising. Third, define a uniform labeling standard modeled on the EU "AI" icon for the affected formats. Fourth, check every chatbot and virtual assistant you operate for a clear disclosure notice. And fifth, document your editorial review process so the exception for humanly reviewed content is demonstrable if needed.

For companies that organize their content production with AI agents, such a labeling workflow can be integrated directly into the existing pipeline - as a fixed part of every publishing step, not as an afterthought. This not only reduces the risk of individual misjudgments but also creates internal clarity about which content needs to be handled how.

In the end, a simple insight remains: the Transparency Code and Article 50 of the AI Act do not require most mid-sized companies to fundamentally reorganize their marketing work. They require clarity about which content is actually affected, a clean editorial review step where it makes sense anyway, and visible, consistent labeling where the obligation applies. Anyone who clarifies this fundamentally now won't have to catch up under time pressure later.