The EU AI Act Starts Labelling Content in August. What Applies to Marketers.

Let me put the disclaimer where you can’t miss it, because this is a compliance topic and I’m a content marketer rather than a lawyer. What follows is my reading of the European Commission’s published guidance as someone who has to operationalise EU AI Act content labelling inside a real content workflow. Treat it as a starting point for the conversation with your legal team rather than a replacement for it. Every source is linked so you can check my reading against the originals.

What happens on 2 August, and what didn’t get delayed

Article 50 of the AI Act sets out transparency obligations, and it applies from 2 August 2026. On 20 July 2026 the Commission adopted its final Guidelines on how those obligations work, running to 51 pages, and Bird & Bird’s read is that national market surveillance authorities can be expected to follow them even though the Guidelines are non-binding and only the Court of Justice can give an authoritative interpretation.

A lot of teams saw headlines about the Digital Omnibus delaying parts of the AI Act and quietly stood down. Sidley’s analysis is that the Article 50 obligations are generally not affected by those proposed delays, with the exception of the Article 50(2) marking obligations that fall on providers. The most broadly applicable section of the entire Act is the one arriving on schedule.

Enforcement sits mainly with national market surveillance authorities, and fines can reach €15 million or 3% of total worldwide turnover for the preceding financial year, with proportionality available for SMEs and small mid-caps.

This reaches you wherever you’re based; the Act works on a market-location principle, so a company outside the EU publishing AI-generated content into the EU market is in scope.

Provider or deployer? You’re almost certainly a deployer

This distinction decides which obligations are yours, and marketing coverage keeps blurring it:

• A provider develops an AI system, or has one developed, and places it on the EU market under its own name or trademark. That’s OpenAI, Anthropic, Google, or your company if you’ve built and shipped an AI product. Providers carry Article 50(1), (2) and (5): designing systems that tell people they’re talking to AI, and marking generative outputs in a machine-readable format.
• A deployer uses an AI system under its own authority in a professional context. That’s your marketing team using a chatbot to draft copy or an image model to produce visuals. Deployers carry Article 50(3) and Article 50(4).

Two details in the Commission’s FAQ matter enormously and I haven’t seen them covered anywhere in marketing press:

1. First, where the deployer is a company, individual employees acting under its instructions are not separate deployers, and the FAQ names content creators and web designers explicitly. The obligation belongs to the organisation.
2. Second, the organisation stays the deployer even when freelancers or contractors operate the system on its behalf, so outsourcing production doesn’t outsource the duty.

Your agency using Midjourney on your brief doesn’t make your agency the responsible party. If the system runs under your authority, the obligation is yours.

The two obligations that actually touch a marketing team

Article 50(4) contains two separate regimes, and conflating them is where most confusion starts.

1. Limb one: deepfakes. This covers AI-generated or manipulated image, audio or video content meeting three cumulative criteria: it resembles a real subject to a high degree, that subject exists or could plausibly have existed, and it would falsely appear authentic or truthful to a viewer. Note what’s absent from that test: there’s no public-interest filter, no marketing exemption, and nothing carving out commercial context.
2. Limb two: text on matters of public interest. This covers AI-generated or manipulated text that is published, informative to the public, and on a matter of public interest. The Commission’s list of qualifying subjects runs to politics and democratic processes, public administration, justice and law enforcement, fundamental rights, public security, public health, environmental protection, consumer safety, and economic, financial, scientific or cultural developments that could be a relevant subject of public debate.

Both limbs require disclosure to the person exposed to the content, at first exposure at the latest, in a clear and distinguishable way.

Which content EU AI Act content labelling actually catches

My working read is below, and this is the section to take to your legal team rather than treat as settled.

1. Generally outside the text obligation. Advertising copy, product descriptions, landing pages, social posts, and newsletters are usually published for commercial purposes rather than to inform the public on matters of public interest. Analysis in the marketing legal press reaches the same conclusion, noting that traditional advertising copy, product descriptions, social media posts and newsletters will generally fall outside Article 50(4)’s second subparagraph. If you write B2B SaaS product pages, most of your text output isn’t the target of this rule.
2. Potentially inside the text obligation. Content that reads as editorial rather than promotional, on a qualifying subject, is where it gets interesting. Native advertising on health or financial topics is the standard example, so is a lot of what the industry calls thought leadership, if it’s published to inform on environmental impact, consumer safety, public health, or economic developments. Given that most thought leadership takes no real position anyway, the compliance question and the quality question point at the same content.
3. The bit marketers are underestimating: imagery. Because the deepfake limb has no public-interest filter, AI-generated visuals that clear the three criteria are caught regardless of commercial context. A photorealistic AI image of a product in a real-looking setting, an AI-generated “customer,” a synthetic office scene presented as your actual office: these are the exposure. The Commission does allow context to matter, since if the intended audience in a specific deployment context doesn’t expect the content to be authentic, it may not falsely appear authentic.
4. A trap worth flagging. The Commission is explicit that deployers can’t rely on the machine-readable marking a provider embeds under Article 50(2) to discharge their own disclosure duty. The invisible watermark your image tool applies satisfies the provider’s obligation while leaving yours untouched, and yours has to be perceivable by a human without special tools.

The editorial control point, which matters more than any other detail

If you read one section, make it this one, because it’s the pressure valve for the whole text obligation and it’s narrower than the summaries suggest.

Article 50(4)’s text obligation doesn’t apply where the content has undergone human review or editorial control and a person or organisation holds editorial responsibility for publication. Human review means deliberate examination of the substance by people with relevant knowledge and professional judgement on the subject matter. Editorial control means a responsible editorial entity with authority to approve, alter or reject the substance on substantive grounds, including fact-checking and assessing source trustworthiness.

And then the line that should worry a lot of content operations: superficial, purely formal, or procedural checks such as spell-checking or grammatical correction don’t count. Greenberg Traurig’s read of the draft guidance was that the Commission interprets this exception narrowly, and the final FAQ language supports that:

A model drafts, a marketer skims for typos, the piece goes live. That workflow doesn’t meet the Commission’s definition of editorial control, and it’s how a great many content teams currently operate.

Read that against how AI actually gets used in most content teams and the risk becomes obvious. The exemption is built for a workflow where a named person with subject knowledge examines the substance, checks the claims, and takes responsibility for publishing it.

That’s the same standard I’ve argued for on quality grounds for years. Documented editorial standards and a real review pass were always what separated content worth publishing from content that merely existed, and the reason good content marketers are editors first hasn’t changed. What’s changed is that the editorial layer now has a regulatory function attached, and review that leaves no trace is worth much less when someone asks you to demonstrate it.

Timelines, and the grace period that you probably don’t have

Three dates govern when EU AI Act content labelling bites, and it’s the middle one where the nuance catches people out:

1. 2 August 2026. Article 50 applies. Deployer obligations under 50(3) and 50(4) start here with no grace period.
2. 2 December 2026. A limited grace period exists, and it’s narrower than the headlines imply. It applies only to AI systems placed on the market before 2 August 2026, and only to the Article 50(2) marking and detection obligation. This is a provider obligation, and if you’re a marketing team, this grace period is almost certainly not for you.
3. Content made before 2 August. It doesn’t need retroactive labelling, though the Commission encourages it. Bird & Bird flags an asymmetry worth knowing: for images, audio and video the relevant date is the date of generation, while for public-interest text the relevant date is the date of publication, so text generated in July and published in September could still need labelling unless the editorial exemption applies.

There’s also a voluntary Code of Practice on Transparency of AI-Generated Content, which the Commission and AI Board have assessed as adequate, sitting alongside the full Guidelines document if you want the primary text.

What I’d do this week

Inventory where AI touches published output, separating text from image, audio and video, because they follow different rules. Flag any content published to inform on a qualifying public-interest subject, including the thought leadership nobody thinks of as editorial. Look hardest at AI-generated imagery, since that’s the exposure with no public-interest filter protecting it.

Then fix the editorial layer, which is the highest-value move available. Name who holds editorial responsibility, define what substantive review means in your workflow, and document that it happened. If you use AI in production, and I use it heavily in specific parts of my own workflow, the review pass is the control point that both protects quality and carries the compliance argument. It’s another version of the argument that the bar for content marketing went up rather than down once production got cheap.

Worth a separate look if you run any customer-facing automation: Article 50(1) puts a disclosure duty on systems that interact directly with people, and the Commission says that explicitly includes agentic AI acting autonomously. If you’ve been building agentic workflows that talk to customers, that’s a different obligation from EU AI Act content labelling and it lands on the same date.

And take it to a lawyer. I’ve linked the Commission’s FAQ, its Guidelines page, and several law firm analyses throughout precisely because you should read the primary sources rather than trusting a marketer’s summary, including this one. The Guidelines are non-binding, the omnibus process is still moving, and I’ve been careful to flag where I’m reading interpretation rather than settled law. Anyone telling you this is simple hasn’t read the 51 pages.

---

This article reflects my reading of published guidance as a content practitioner, current as of late July 2026. It isn’t legal advice, the Guidelines are non-binding, and the regulatory picture is still moving. Verify against the primary sources linked above and take advice on your own situation.