Nineteen Days to Article 50: What the EU AI Act's Transparency Rules Mean for PR

In nineteen days, the transparency provisions of the EU AI Act become applicable. Per the European Commission's official guidelines on the AI Act's transparency obligations, Article 50 applies from August 2, 2026 — and while most coverage has framed the milestone as a technology-law story, its most immediate practical impact lands on content, communications and PR operations.

What Article 50 requires

• Synthetic content must be marked. Providers of generative systems that produce text, images, audio or video must ensure outputs are machine-readably marked as artificially generated or manipulated.
• Deepfakes must be disclosed. Anyone deploying systems that generate or manipulate image, audio or video content must disclose the artificial nature of the output.
• Public-interest text must be labeled. AI-generated text published to inform the public on matters of public interest must be disclosed as AI-generated — unless a human has reviewed and edited it meaningfully, in which case the human editor takes responsibility.
• Chatbots must identify themselves. Systems interacting directly with people must be designed so users know they are talking to AI.

Why PR is in scope

The public-interest text provision reads like it was drafted for the communications industry. Press releases, corporate statements, executive thought-leadership and newsroom content exist precisely to inform the public — and the Act's carve-out creates a bright operational rule: meaningful human review converts a disclosure obligation into an editorial responsibility. Teams therefore need documented review workflows, not just style guidance. Who reviewed, what was changed, and on what date become compliance facts.

Reach is the second landmine. The obligations apply to providers and deployers whose output reaches EU users, regardless of where the company sits — a US, UK or Asian comms operation publishing English-language content for European audiences cannot assume its home jurisdiction insulates it.

Anatomy of a review workflow

The meaningful-review carve-out is doing a lot of work, and it deserves to be engineered rather than assumed. A workflow that survives scrutiny has four properties. It is named: a specific human editor signs each public-interest piece, not a team alias. It is substantive: the reviewer verifies claims, sources and quotes, which is editing in the journalistic sense rather than proofreading. It is timestamped: the record shows review happened before publication. And it is versioned: the draft and the reviewed text are both retained, so the difference between them is visible.

None of this requires new tooling for most teams — a CMS with revision history and an approval step covers it. What it requires is that the approval step actually gates publication, rather than being a formality clicked after the fact. The gap between those two designs is the gap between invoking the carve-out and merely hoping for it.

Related stories: AI Adoption in PR Has Leveled Off — Muck Rack's 2026 Report Shows the Real Divide Is Now Training · Omnicom Closes $9B IPG Deal — and PR's Biggest Names End Up Under One Roof.

The agency question

Most public-interest content in the PR industry passes through agencies, freelancers or regional consultancies before publication, and Article 50 does not care. The deployer publishing to EU audiences holds the obligation, and the deployer's contracts decide who actually performs the work. Agency agreements drafted before 2026 almost never allocate AI-transparency duties, because the duties did not exist. The uncomfortable conversation to have before August 2 is contractual: who documents review, who warrants that machine marking survived the production chain, and who bears the loss if a regulator disagrees.

Agencies should welcome that conversation rather than resist it, because it is also a differentiator. A shop that can show a labeled, reviewed, logged pipeline for AI-assisted content is selling compliance as part of the retainer. A shop that answers "we use AI responsibly" has sold a slogan.

What the obligation does not cover

Scope discipline matters as much as scope panic. Article 50 governs transparency about how content was made, not whether AI may be used. There is no duty to disclose that a press release was drafted with AI assistance and then substantially written by a human; the public-interest provision attaches to AI-generated text published as such, and the meaningful-review carve-out exists precisely to normalize assisted drafting. Internal drafts, background documents, and content that never informs the public sit outside the labeling duty. Teams that read the Act as a ban on AI in communications will over-comply, spend on the wrong thing, and still miss the actual exposures — published synthetic media and unreviewed public text.

The line the Act draws is also a useful editorial line. Content that would embarrass the organization if labeled as machine-made is content a human should have reviewed anyway. In that sense the compliance standard and the quality standard converge: the review workflow that satisfies a regulator is close to the one that satisfies a demanding editor. Organizations that already run real editorial review will find the new duty nearly free; organizations that never did will discover the gap under time pressure, which is the expensive way.

What to do before August 2

Per the Commission's accompanying work, a voluntary Code of Practice on Transparency of AI-generated Content is being developed to operationalize labeling — but the underlying obligations do not wait for it. A minimum pre-deadline checklist: inventory where generative AI touches your content pipeline; decide the human-review standard for public-interest text and document it; ensure your agency contracts allocate labeling responsibility; and brief spokespeople that "disclosed AI-assisted" is becoming a credibility signal rather than a confession — early transparent adopters are likely to fare better with journalists already sensitized to synthetic content.

Nineteen days is enough time for the checklist. It is not enough time to invent a workflow, staff it and backfill documentation — which is the honest argument for starting today rather than testing what "applicable" means.

Per European Commission guidelines on AI Act transparency obligations; Article 50 applicable from August 2, 2026.