AI is already present on many websites, often without anyone treating it as a separate operational topic. A chatbot answers questions. Images are edited with AI. Copy is drafted with model assistance. Support, video and audio tools add more moving parts.
Each tool can look manageable on its own. The gaps appear between them: who knows which uses customers actually see? Who reviews them? Who decides whether a notice or label is needed?
Those questions need a clear owner.
What changes on 2 August 2026
The European Commission has announced that further EU AI rules will be enforced from 2 August 2026. For certain AI systems, it identifies additional transparency duties.
The communication covers, among other things, three different situations:
- In certain AI interactions, people may need to be told that they are speaking with an AI system.
- Deepfakes must be labelled.
- Certain AI-generated or altered content may need a machine-readable marking.
This does not mean that every AI-assisted item on every website needs the same label. The answer depends on the actual use, the organisation’s role, the system and the type of content. It needs to be assessed case by case.
The enforcement date also does not mean that every obligation in the wider AI framework starts at once. A dependable decision needs the current legal text, official guidance and the details of the implementation.
Three questions that need separate answers
1. Is a person speaking with a system?
When a visitor interacts with a chatbot or another AI system, a clear notice may be needed. That is about the interaction itself. It is different from labelling an image or video.
2. Has realistic content been generated or altered?
For deepfakes, the focus is on labelling artificially generated or manipulated content. That is not the same assessment as using text assistance inside an internal editorial workflow.
3. Does the origin need to be technically detectable?
Certain AI-generated or altered content may require a machine-readable marking. A visible notice and a technical mark are not automatically the same thing.
Keeping these cases separate removes a lot of noise. Each customer-facing use can be checked against the question that actually applies to it.
Where to start
The work does not begin with another website banner. It begins with an honest overview.
1. Record the use cases
List where AI appears in customer communication or content production: chatbots, generated or edited images, video and audio, automated replies, editorial workflows and embedded provider services. A small tool belongs on the list when customers can see its output.
2. Give each use case an owner
Every use case needs a named person or role. That person does not have to resolve every legal question alone. They should know what is in use, which information is available and when a new review is needed.
3. Define what triggers another review
Review is not a one-off task. A new provider, feature, content type or official guidance can trigger another review. Put those triggers into editorial review, release management or procurement.
4. Record the decision
For each use case, record the purpose, system or provider, affected content, responsible role, transparency option considered, open questions and next review date. This record is not legal advice. It makes the required review easier to trace.
What happens without clear ownership
Gaps appear. The chatbot is disclosed, but an edited image is not. A technical marking disappears during export and no one notices. Later, the team has to start from scratch: which tools were active, who decided, and which assets are affected?
That costs time and creates uncertainty. More importantly, no one is clearly responsible for starting the next review.
Why the website is often where this comes together
AI features rarely sit in one place. They can be part of the website, CMS, support stack, campaigns or an external provider. Technical delivery, editorial decisions and accountability are then spread across several people.
At DKSIGN, we look beyond one website feature. We review how the connected workflows fit together: where content is created, where it changes, who reviews it, what customers see, and what happens when the workflow changes later.
That is technical responsibility and maintainability, not legal certification. The legal interpretation of a specific case belongs in an appropriate legal review.
A practical next step
If your team lacks a clear view of the AI features connected to your website, we can review the functions in use, clarify ownership and agree the next steps.
*Note: This article is editorial information, not legal advice or a compliance determination. Requirements may differ by system, role, content and use case.*