EU AI Transparency Rules Are Live. Is Your Site Ready?

For a while, AI compliance sounded like a problem for companies with a legal department and a thousand employees. That changed on August 2. The transparency obligations in Article 50 of the EU AI Act now apply, and they touch some very ordinary things: a chatbot on a website, an AI-generated image in a campaign, or automatically published text about a public-interest topic.

The useful way to read this is not "we need to put a legal disclaimer everywhere." It is: people should be able to tell when software is talking to them or when material was made by a machine. The European Commission's guidance is specific on that point. It requires transparency when people interact directly with an AI system and puts separate rules around identifiable synthetic content, deepfakes, and certain public-interest text without human editorial review.

The website question is simpler than it looks

Most small businesses do not build foundation models. They use tools somebody else built. That does not make transparency someone else's problem.

If a visitor opens a support chat and assumes a human will answer, the interface needs to correct that assumption. If an assistant can take an action, such as booking an appointment or changing an order, the user needs to understand the boundary between an automated suggestion and a confirmed outcome. If a campaign uses a realistic generated voice or image, the label needs to be visible enough to matter, not hidden in a footer.

That is a product-design question before it is a legal one. A clear line such as "You are chatting with our AI assistant" is usually better for trust than a vague human-sounding name, even if the latter produces more initial engagement. When someone discovers the truth later, the short-term conversion trick becomes a long-term credibility cost.

A practical checklist for an AI feature

Before adding an AI widget to a website or internal tool, we would ask five questions:

1. Can a person tell they are interacting with AI? The disclosure should sit next to the interaction, not only in terms and conditions.
2. What can the system actually do? Drafting a reply, searching a knowledge base, and issuing a refund are different risk levels. The UI should not blur them together.
3. Where does a human review the output? Human review is especially important for public claims, pricing, advice, or content that could affect someone materially.
4. Can the team trace what happened? A useful business system keeps enough context to investigate a wrong answer without turning every customer interaction into a surveillance project.
5. Who owns the content and the disclosure? A vendor may provide a model, but the business deploying the experience still owns the customer relationship.

The Commission also notes a grace period until December 2026 for some marking obligations on generative systems already placed on the market before August 2. That is not a reason to wait. It is time to make a deliberate inventory: which AI tools are visible to customers, what do they say, and who has checked the experience end to end.

The opportunity is better interfaces, not more warnings

The businesses that handle this well will make AI feel less mysterious. A helpful assistant can state what it knows, what it cannot do, and when it is handing a question to a person. That is a stronger interaction than pretending every answer came from an employee.

This is also why generic AI widgets are rarely the final answer. A useful assistant needs the business's real context: current services, policies, availability, language, and escalation paths. At Legacies, when we build an application or website workflow with AI in it, the work is not just connecting a model to a chat bubble. It is deciding the job, permissions, handoff, and disclosure so the tool saves time without making customers guess what is happening.

Start with the visible parts

Do not begin with a giant compliance programme. Open your site as a first-time visitor. Try the chat, automated email, booking flow, and any generated media. If a reasonable person could mistake an automated interaction for a human one, fix that first.

Then document the workflow in plain language. This is valuable beyond regulation: it exposes whether the automation is actually useful. If nobody on the team can explain what an agent is allowed to do, it is probably not ready to act on a customer's behalf.

The AI Act's transparency rules are ultimately about preserving informed choice. For small teams, that is a good product principle anyway. Build systems that are clear enough to trust, and customers are much more likely to keep using them.