In B2B marketing, a typical content workflow looks like this: The marketing or comms agency drafts a piece of copy, whitepaper or press release. They do so using AI. Then someone makes small adjustments or scans it with one eye half closed before it is approved and the piece goes online.
Earlier this year, that content workflow was a potential quality problem. Since August 2, it is also a legal question, given Article 50 of the EU AI Act. (You can learn more about the guidelines here: The European Commission.)
When the law went into effect, most of the press coverage was about deepfakes. The part that matters for anyone running a marketing or communications team is less easy to spot, and it sits in Article 50(4). It is a particular exemption: Publish AI-generated text to inform the public on a matter of public interest, and you have to say so, unless that text went through meaningful human review and editorial.
Europe has drawn a legal line between text a human being actually authored and text a human being merely watched go past. The first kind needs no label. The second does.
In my view, that distinction is the most useful thing a regulator has done for content quality in a decade.
Editorial standards
The Commission is unusually blunt about what doesn’t qualify. Review has to be a deliberate examination of the substance of the content, by someone with relevant competence and professional judgement. Spell-checking doesn’t count as a review. Grammar fixes don’t count. Cursory approval without substantive engagement explicitly does not count.
Spell-checking doesn’t count as a review.
The EU basically created an editorial standard, and it’s the one your best editor on staff already uses. In other words, human writers and editors matter, and human judgement must be the last word.
Human judgement must be the last word.
What type of content falls under the law?
If you’re a niche content creator, this does not exempt you from the law and you should definitely seek appropriate legal counsel (which this blog is not).
A trade blog read by 900 procurement engineers is published content and of the public interest, same as a newspaper. Therefore, it is likely to fall under the law. And emissions targets are public interest, even if only 40 people read them. Those texts have to be labeled.
More examples: If you sell fans and motors, your content is about energy consumption. If you sell filtration and ventilation, it is about air quality and health. If you sell building products, chemicals or automation systems, this impacts emissions, water and workplace safety. I’d say all of these count.
However, if you’re writing industrial product descriptions, like the characteristics of materials and parts, this is not public interest, as long as there are no claims related to health or safety. These texts likely do not fall under the law.
The test I’d use
To gauge whether your editorial processes are compliant or non-compliant, set the new EU law aside for a moment, and try this instead.
Ask around among your marketing colleagues: ‘Who wrote that big piece published last month?’ If you get a name, and that person can tell you what they changed and why they changed it, you’re probably fine. If you get a shrug and no one can think of a name, you may have an issue, considering the new law.
It’s not uncommon for unchecked work to get published.
Muck Rack’s State of AI in PR 2026, a survey of 564 PR professionals done in January of this year, found that 76% now use generative AI while just 51% work somewhere with a policy governing it.
That gap is a quality risk. On August 2, it became a legal risk, too, for businesses operating in Europe.
In my view, the new law is good for business and good for society.
Communications are about trust. If no human is reviewing materials that go out, that’s a breach of trust.
That’s just what this law is trying to change.