Do I Need to Tell Visitors They Are Talking to AI?

You should clearly tell visitors when they are interacting with an AI assistant. In the EU, Article 50(1) of the AI Act requires providers of relevant systems to ensure that people are informed, unless the AI interaction is obvious in context. These transparency obligations apply from 2 August 2026. Other jurisdictions have their own rules, so treat clear disclosure as a practical baseline rather than assuming one law covers every visitor.

For a Niwa website assistant, the useful message is simple: identify it as AI, explain the customer tasks it handles and show how to reach the business. A visitor should understand the conversation before sharing information or relying on an answer.

What the EU rule requires

The European Commission's Article 50 guidance explains that the relevant interaction is a direct exchange between AI and a person. It also says the notice must be clear, distinguishable and accessible from the start of the first interaction.

The legal responsibility under Article 50(1) is framed as a provider duty to design and develop the system appropriately. A business using a supplier's tool should check the supplier's implementation and preserve the notice in its own configuration. If the business develops or has a system developed under its own brand, it may also have provider responsibilities.

The exception for an interaction that is already obvious is contextual. A familiar-looking chat window, a friendly first name or a small robot icon does not give you a reliable basis for assuming every customer understands the system. An explicit label is usually easier to test than an argument about what customers ought to infer.

This notice is different from the separate rules about machine-readable marking of generated outputs. Writing "AI assistant" in a chat does not, by itself, satisfy every AI Act requirement that might apply to the provider.

Put the explanation where the visitor starts

Consider the path from the closed launcher to the first reply. Someone arriving on a phone may open the assistant without reading the page footer or privacy policy.

Use a short visible label in the chat heading and an introduction in the opening message. Keep it readable after the visitor opens the keyboard. If the assistant speaks aloud, consider how a user receives the same information through audio and an accessible interface.

An example for a shop is:

"I'm the AI assistant for Oak Street Shoes. I can help with product information, sizing guides and delivery questions. For help from our team, use the contact link below."

That is an illustrative message for a fictional shop. Adapt the scope to your actual business. Do not promise order changes, refunds or specialist advice unless the assistant can legitimately support those tasks.

Avoid names and pictures that imply a person

You can give an assistant a friendly name, but keep its AI identity visible. "Maya, AI assistant" is clearer than "Maya from customer support" beside a portrait that appears to show an employee.

Also check the messages generated during handoff. A statement such as "I'll connect you with the team" should lead to a real route. A statement such as "A person is reviewing this now" needs an actual person reviewing it.

If your team takes over, tell the visitor that the conversation has changed. If the business only receives an email request, explain that outcome and your normal response arrangements. Clear expectations matter as much as a label.

Disclosure, privacy information and consent do different jobs

The AI disclosure explains who or what is communicating. A privacy notice explains how personal data is used. Consent, when it is the appropriate legal basis or a marketing requirement, records a person's choice for a specific purpose.

Under GDPR Article 13, applicable privacy information must be provided when collecting personal data. A short chat introduction can link to the full notice. It should not pretend that "you are talking to AI" explains retention, suppliers or international transfers.

Likewise, a person opening chat has not necessarily agreed to a newsletter or future WhatsApp promotions. Give those optional purposes their own appropriate choice instead of folding them into the AI notice.

Check the disclosure in every supported language

Translate the meaning, including the task boundary and contact route. Keep the AI identity visible when the conversation language changes. A page translated into Portuguese with an English-only disclosure can leave the visitor with an unclear explanation.

The Niwa setup guidance is a starting point for configuring the website assistant. Review the actual introduction, labels and language behaviour in your installation. This is a verification task, not a claim that selecting a plugin automatically satisfies every disclosure requirement.

A useful acceptance check is to ask someone who has not seen the setup to open chat and explain what they think they are using. Do this on a phone as well as a desktop. Check keyboard access and screen-reader output, and confirm that the contact and privacy links work.

Choose a clear introduction for your business

For a salon, the assistant might explain services and direct visitors into the website booking flow. For a retailer, it might answer catalogue and delivery questions. Write the introduction around that concrete role.

Existing WordPress owners can use the standalone Niwa plugin. If you need the website work too, Niwa Business is USD 150 for a custom WordPress site or migration, hosting, maintenance and AI adaptation.

Send Niwa your website address and intended chat task. Include the languages your visitors use so the introduction, support route and customer journey can be discussed together.

Legal references checked on 9 October 2026. The EU rules above do not replace an assessment of your own jurisdiction or provider role.