Do we have to tell customers they are talking to an AI?

In most jurisdictions with AI-specific regulation, yes, businesses are required or strongly expected to disclose that a customer is interacting with an AI system rather than a human, and this obligation is becoming more explicit as regulation catches up with the technology. The EU AI Act, which entered into force on 1 August 2024, includes a transparency obligation requiring that people be informed when they are interacting with an AI system unless it is obvious from the circumstances, with the bulk of the Act's provisions, including this one, applying from 2 August 2026. Several US states and other jurisdictions have separate bot-disclosure laws, particularly around sales and commercial communications, so requirements vary by where your customers are located rather than where your company is based. Beyond legal compliance, disclosure is generally good practice for trust: customers who later discover they were misled about talking to a bot tend to react more negatively than those told upfront, especially once the AI hands off to a human. A simple, consistent disclosure at the start of a chat or call, without over-explaining, satisfies most requirements without disrupting the conversation. Nanobase AI builds this disclosure into the conversation design by default rather than treating it as an optional setting.

Disclosure is a design decision, not a checkbox

Treating AI disclosure as a legal checkbox to satisfy once, rather than a conversation design choice, produces the two failure modes companies actually run into: either a buried disclaimer nobody reads, or an over-explained intro that makes customers wait through a paragraph before asking their question. A single, persistent signal, a labeled avatar and name in chat, or one short sentence at the start of a voice call, satisfies most disclosure obligations without disrupting the interaction that follows. The pattern that works best keeps the disclosure visible throughout the session rather than only at the start, since a customer who joins a chat mid-conversation or gets transferred between channels should not have to guess whether they are still talking to the same type of agent.

The regulatory picture varies by customer location, not company location

Because obligations attach to where the customer is, a multinational support operation generally ends up applying the strictest applicable rule everywhere rather than maintaining a jurisdiction-by-jurisdiction disclosure toggle, which is simpler to build and audit.

Getting the human handoff moment right

Disclosure gets harder, not easier, at the transition point between AI and a human agent, and this is where poorly designed systems create the most customer frustration. A clean handoff states plainly that the conversation is moving to a human agent, ideally by name, rather than letting the customer discover the switch mid-sentence when the tone or response speed suddenly changes. The reverse handoff, a human bringing an AI assistant back into a conversation for a specific task like pulling up an order, deserves the same clarity, since silent switching in either direction is what erodes trust more than the presence of AI itself. Voice deployments in particular benefit from a distinct tone or brief audio cue at the handoff point, since customers cannot see a UI label change the way chat users can.

Designing against disclosure fatigue

The goal is satisfying the obligation with the least possible interruption to the conversation, which a few consistent habits achieve without any legal ambiguity.

1. State the AI disclosure once, clearly, at the start of the interaction rather than repeating it every turn.
2. Keep a persistent, low-key visual or verbal marker, such as a labeled avatar, active for the length of the session.
3. Reintroduce the disclosure explicitly only at a channel switch or a human handoff, not at every message.
4. Avoid giving the AI a human name or first-person claims of being a person, which creates a disclosure contradiction customers notice quickly.

Frequently asked questions

Does a chat bubble labeled "AI Assistant" satisfy EU AI Act transparency requirements?

A clearly labeled, persistently visible indicator generally satisfies the spirit of the obligation, though the exact compliance bar depends on how "obvious from the circumstances" is interpreted in your sector, so pairing the label with a brief opening statement is the safer approach.

Do voice agents need a spoken disclosure at the start of every call?

Yes, since a customer cannot see a chat label on a phone call, a short spoken line at the start of the call is the equivalent control and is generally the more defensible approach for voice deployments.

Should disclosure differ for internal employee-facing bots versus external customers?

Internal tools generally carry lower disclosure risk since employees can be informed through onboarding and policy, but any tool interacting with external customers should not rely on internal-only communication to satisfy disclosure obligations.

What happens if we don't disclose and a customer finds out later?

Beyond the direct regulatory exposure, undisclosed AI interactions that customers discover after the fact tend to damage trust more than the AI's involvement would have on its own, which is a real cost even in jurisdictions with limited enforcement history.

How Nanobase AI helps

Nanobase AI, a Silicon Valley enterprise AI engineering company, builds disclosure directly into conversation design across chat and voice channels, including the handoff moments where most implementations fall short. This work is usually delivered as part of a broader AI agents and process automation engagement rather than a standalone compliance add-on.

Ready to discuss your project? Contact Nanobase AI or email hello@bumu.tech.