Your Chatbot Needs to Say It's a Bot

The EU AI Act's transparency obligations (Article 50) kick in August 2, 2026. If your chatbot interacts with people in the EU, it must clearly disclose that they're talking to an AI system. Not buried in a terms of service page. In the interaction itself.

Fines for non-compliance with transparency obligations: up to 15 million euros or 3% of global annual turnover, whichever is higher.

Even if you're not in the EU, this matters. If any of your customers are EU residents, the regulation applies to you. And other jurisdictions are following suit.

What You Need to Disclose

The AI Act requires that people "are informed that they are interacting with an AI system, unless this is obvious from the circumstances and the context of use." The law also requires that AI-generated content be marked in a machine-readable format.

In practical terms, this means:

Your chatbot needs to clearly state it's AI-powered before or at the start of a conversation. A message like "Hi! I'm an AI assistant. I can help with common questions, and I'll connect you with a person if I can't." works.

The disclosure needs to be visible, not hidden. Putting "AI-powered" in size-8 font at the bottom of the widget probably won't satisfy regulators. The disclosure should be part of the natural conversation flow.

If your AI generates or manipulates images, audio, or video, you also need to label that content as AI-generated. This mostly applies to deepfakes and synthetic media, but if your support bot generates visual content, it applies to you too.

Machine-readable marking means your AI-generated responses should include metadata indicating they were generated by AI. This is still being defined in technical standards, but adding a data attribute or API header is the likely implementation.

What You Don't Need to Disclose

You don't need to explain how your AI works. No need to describe your model architecture, training data, or classification methodology.

You don't need to disclose AI involvement in internal routing or AI-assisted drafting. If AI classifies a message and routes it to a human agent who then responds, the customer is talking to a human. Similarly, if a human agent uses AI to draft a response but reviews and sends it themselves, most interpretations say that's a human response. The disclosure requirement targets direct AI-to-customer interactions.

Sample Disclosure Text

Here are a few approaches, from minimal to thorough:

Minimal: "You're chatting with our AI assistant."

Standard: "I'm an AI assistant trained to help with common questions. I'll transfer you to a human agent for anything I can't handle."

Thorough: "This is an automated AI assistant. I can help with order status, returns, account questions, and general information. For complex issues, I'll connect you with our support team. Response times for human agents vary."

Widget-level: Add a persistent "Powered by AI" badge to the chat header so the disclosure is always visible, not just in the first message.

US Requirements

The US doesn't have a federal AI disclosure law (yet). But several states have rules:

The California Consumer Privacy Act (CCPA) has been interpreted to require disclosure when AI makes decisions about consumers, and California has additional proposed AI transparency legislation in progress. New York City's Local Law 144 requires disclosure for AI in hiring. More states are adding AI transparency requirements.

Best practice: just disclose everywhere. The regulatory direction is clear. Every major jurisdiction is moving toward mandatory AI transparency. Building disclosure into your chatbot now saves you from retrofitting later.

Implementation

Most chatbot platforms (including Supp's widget) let you customize the greeting message. Change your default greeting to include AI disclosure. If you're building a custom integration, add the disclosure to the first message in every conversation.

For the machine-readable requirement, add structured data to your widget's HTML. A simple approach is a meta tag or data attribute on the chat container element indicating AI involvement. The specific technical standard is still being finalized by the EU AI Office, but having something in place shows good faith.

Test that the disclosure is visible on mobile. Chat widgets often render differently on small screens, and a disclosure that's clear on desktop might get cut off on a phone.