EU Sets 2 Aug 2026 AI Labelling Deadline
The European Commission has published its Guidelines on transparency obligations for providers and deployers of certain AI systems, and it comes with a hard date. Article 50 of the AI Act applies from 2 August 2026. From that day, AI-generated content will need machine-readable marks, and people must be told plainly whenever they are interacting with an AI system rather than a human.
The guidelines were last updated on 6 August 2026 and were built with input from Member States, the AI Board and a public consultation. If you produce content, run chatbots, or publish anything touched by generative AI, this is now an operational requirement, not a policy footnote.
What actually happened
The Commission has clarified who must do what under Article 50. It splits duties between two roles: providers (the ones who build and supply AI systems) and deployers (the ones who use them in the wild). Most marketing teams sit firmly in the deployer camp, and that distinction decides your obligations.
The stated reason is blunt. The Commission says generative and interactive AI is making it harder to tell AI content from authentic, human-created content, raising risks of misinformation, fraud, impersonation and consumer deception at scale. The fix is disclosure.
When it bites, and over what timeline
The single date to circle is 2 August 2026. That is when the transparency obligations under Article 50 apply. The guidelines and the accompanying Code of Practice on Transparency of AI-generated Content are already live, so there is no reason to wait until the summer to start.
In my opinion, treating August as a start line rather than a finish line is the mistake teams will make. Labelling and disclosure need to be baked into workflows, and that takes longer than a compliance memo.
How the obligations work
The rules divide cleanly. Here is the split.
| Role | What you must do under Article 50 |
|---|---|
| Providers (AI system builders) | Design systems so individuals are explicitly told when they interact with AI directly. Add machine-readable marks so AI-generated or manipulated content can be detected. |
| Deployers (businesses using AI) | Inform individuals when exposed to emotion recognition or biometric categorisation, when shown deepfakes, and when reading text on matters of public interest published without human review or editorial control. |
Read that deployer list carefully. The line that matters most for content teams is the third one: text on matters of public interest published without human review or editorial control must be disclosed. If you are auto-generating commentary on news, health, politics or anything of public importance and pushing it live without a human editor, you now owe your reader a label.
What this means for a marketing or GEO team
From my observation, most of the exposure here is not in your chatbots, it is in your content pipeline. The good news is the guidelines carve out exceptions, including standard editing. AI used to tidy grammar or format is not the target. AI used to write and publish public-interest content unchecked is.
Here is the practical path I would take:
- Map where AI touches published output. Chatbots, on-site assistants, auto-generated articles, synthetic images and video. List every surface.
- Add clear "you are talking to an AI" disclosure to any interactive assistant. This is the easiest win and it is non-negotiable for direct interactions.
- Insert genuine human review into any workflow producing public-interest text, and keep a record of it. Editorial control is your defence.
- Adopt machine-readable marking for synthetic content. The Commission says adhering to the Code of Practice is one way to demonstrate compliance; if you skip the code, you must prove equivalence by other adequate means.
Who enforces it
Enforcement sits with the national market surveillance authorities, the AI Office for systems under its supervision, and the European Data Protection Supervisor when EU institutions are the providers or deployers. This is not a voluntary standard, it is supervised law.
The bigger shift for discovery
I think this is really about trust becoming a labelled, machine-readable signal. As AI assistants become the front door to discovery, provenance and disclosure start to shape what gets surfaced and cited. This echoes what we have already seen with provenance and verification becoming rank currency, and it lands on the same measurement problem teams are wrestling with as they try to make AI referral traffic countable in GA4.
The action to take now
Do not wait for August. This week, audit every place AI produces or shapes content that reaches a customer, and decide two things for each: does it need a "this is AI" disclosure, and does it need a human editor in the loop. You can download the full guidelines from the Commission to check where your use cases fall. Trust is becoming compliance, and disclosure is becoming a feature. Build it in now.
Tags