Take one illustrative line from an impact section: ‘We will reach 40,000 users.’

→ The university means people reached through communication.
→ The SME means potential customers.
→ The public authority means citizens expected to benefit from a service.

One number. Three different commitments. The wording looks consistent. The expectations are not. AI did not create the gap. It made it look finished. Before anyone validates the figure, someone has to get the partners to agree on what it measures.

Communication, dissemination and exploitation make that distinction essential: building awareness, sharing research results and putting them to use require different measures of success. An audience reached is not proof of adoption. A credible proposal explains who is expected to use the results, what would enable that use and which partner takes responsibility for the next step.

Resolving this takes negotiation from the start. Universities, companies and public bodies bring different missions, data policies and definitions of success. Alongside budgets and work packages, partners need to agree what information may enter AI tools, who validates claims against sources and assumptions, and who approves the commitments made in their name. Only then does the writing team know what it is writing.

Getting institutions to agree on what they promise, how they will demonstrate delivery and who takes responsibility has been central to my work. That agreement gives a proposal its credibility long before submission. Building it is what I want to bring to a research office or an institution with European partnerships at its core. For coordinators and research offices: who checks that a shared target means the same thing to every partner?

Since 2 August, Article 50 of the AI Act has applied new transparency obligations to certain systems and to content generated or manipulated by AI. But the answer is not simply that “all AI-generated content must carry a label.”

The regulation distinguishes between providers and deployers of AI systems, between types of content, and between contexts of use. Providers of generative systems have machine-readable marking obligations; deployers must inform people in specific cases, such as deepfakes or certain texts on matters of public interest.

There is also a transitional period, until 2 December, for certain marking obligations that apply to generative systems already on the market before 2 August.

It looks like a legal detail. But I believe it reveals something much larger.

Between adopting a regulation and getting an organisation, or a society, to know how to apply it, there is a vast governance space.

And that space is not closed by regulatory compliance alone.

I am currently involved in a consultancy linked to an audit on AI governance and AI Act implementation for international organisations within and beyond Europe. One issue comes up again and again: complying with a rule, and getting an organisation to understand it, build it into its processes and be able to explain it, are different capabilities.

A policy can be approved and still not have become a procedure.
A procedure can exist and still not have reached the people who make decisions.
A transparency obligation can be in force and still be unknown to those it is meant to protect.

This is one of the major challenges of the next phase of AI in Europe: closing the gap between regulation and implementation.

The Regulation itself explains that these obligations aim, among other things, to enable people to recognise when they are interacting with AI or are exposed to certain AI-generated content, and to make informed decisions.

That is why communication does not come at the end of governance. It is part of it.

Europe has spent recent years building the rules for AI. The imminent challenge will be to turn those rules into behaviours, processes and decisions that organisations and citizens can actually understand and use.

Governance begins with the rule. But it only works when it reaches the person.

Where do you think the biggest gap lies today: in knowing the regulation, interpreting it correctly, or turning it into real processes within organisations?

Original (Spanish): europeancommunicationstrategies.com, 17 September 2026