Do you have to disclose that an AI is answering your calls?

If you’re considering putting an AI voice agent on your company’s phone, this is the question worth settling before the first call and not after: yes, you have to disclose it. European regulation requires telling the person that they’re talking to an automated system. What almost nobody explains is how to say it without scaring off the caller, and who’s responsible if one day it wasn’t said.

This is a practical explanation, not legal advice. For a specific case, check with your legal adviser.

What the rule says, in short

The European Artificial Intelligence Regulation (the AI Act) sets transparency obligations for AI systems that interact directly with people. The underlying idea is simple: nobody should believe they’re talking to a human when they’re talking to a machine.

That means the person calling must be able to know they’re being served by an automated system, unless it’s obvious from the context. In a well-made voice call it’s never obvious, so in practice you have to say it.

On top of that comes GDPR, which governs recordings. If the agent records the conversation or stores the transcript, you’re processing personal data and you need to inform people of that, have a legal basis, and have a retention policy.

They’re two separate obligations and it’s best not to mix them up: one is “you’re being served by an AI,” the other is “this call is being recorded.” You can have the first without the second, but if you record, you need both.

Where it gets complicated: the notice kills the conversation

Here’s the real problem, and it’s a business one, not a legal one. The notice goes at the start of the call, which is exactly the most fragile moment. If you open with a paragraph of legal terms, two things happen: people hang up, or they tune out and then don’t hear the first question.

The solution isn’t hiding the notice. It’s writing it the way a person talks, not the way a legal department writes.

How not to say it

“We hereby inform you that this communication is being handled by an artificial intelligence system and may be recorded in accordance with current data protection regulations, and you may exercise your rights…”

Nobody listens to the whole thing. And worst of all: it sounds like a bad call center before you’ve said anything useful.

How to say it

“Hi, I’m [company]’s virtual assistant. I’ll be helping you, and if you’d prefer, I can put you through to a person. How can I help?”

It complies: it says it’s a virtual assistant, in the first sentence and without ambiguity. And it also does something the long notice doesn’t: it offers the human exit right away, which is what truly reassures the caller.

If you also record, it’s added briefly after the first response, not chained to the greeting: “Just so you know, we record the call to review the quality of the service.”

The three rules that make it work

• First, and brief. In the opening sentence. If you leave it until the end, it’s no longer transparency, it’s a confession.
• Offer a person at the same moment. The caller’s fear isn’t the AI: it’s getting trapped in it. If you tell them from the start they can get out, most won’t even ask.
• With the same voice and the same tone. A prerecorded, metallic notice followed by a natural voice sounds like a trick and breeds distrust.

And if it isn’t disclosed, who’s responsible?

This is the question a business owner really asks, and the uncomfortable answer is: you.

In front of your customer and in front of the authority, the one providing the service is your company. The technology provider gives you the tool, but the one who decides what the agent says when it picks up, and who answers for customer service, is you.

That changes something when you’re choosing a provider: the notice can’t be a box someone on your team has to remember to configure. It should come set up by default, reviewed, and set up again whenever the script changes. If the provider leaves it as your task, the risk is yours too.

There’s a detail that often slips through: when you change the agent’s script months later, the notice can fall out without anyone noticing. That’s why it’s part of the periodic review, not the launch.

Checklist before opening the phone to AI

1. The agent says in its first sentence that it’s a virtual assistant.
2. It offers to transfer to a person from the start, and actually does so when asked.
3. If it records or transcribes, it says so and there’s a legal basis for it.
4. You know in which country the recordings are stored and how long they’re kept.
5. Your customers’ conversations aren’t used to train third-party models, or you’ve expressly decided they can be.
6. Your privacy policy mentions automated phone handling.
7. The notice is included in the periodic script review, not just the initial version.

What nobody tells you: disclosure sells

Intuition says hiding the AI gives better results. Experience says the opposite.

When someone discovers midway through a conversation that they’ve been talking to a machine, they feel deceived, and your brand takes the hit. When they know from the start, they adjust their expectations: they get to the point, speak more clearly, and appreciate getting an answer at eleven at night instead of an answering machine.

Put another way: the notice isn’t the toll for using AI. It’s what makes using it not cost you customers.

How we handle it at Glofera

In the voice agents we manage, the notice comes set up by default and adapted to each company’s tone, with the exit to a person active from the first sentence. And it’s reviewed every time the script is touched, which is when problems tend to appear.

If you’re thinking about taking the step, start with the free communications analysis: we look at how your calls come in today and which part makes sense to automate. And if you want to understand first the differences between building it yourself or having it managed for you, it’s explained in managed voice agent versus the one you build yourself.

Frequently asked questions

Is it mandatory to disclose that an AI is handling a call?

Yes. The European Artificial Intelligence Regulation sets transparency obligations for systems that interact directly with people: the other party must be able to know they’re talking to an automated system, unless it’s obvious from the context, which isn’t the case in a voice call.

At what point in the call do you have to say it?

At the start, in the opening sentence. A notice at the end of the conversation doesn’t serve the transparency purpose and also creates a feeling of deception for the caller.

Is disclosing that it’s an AI the same as disclosing that the call is recorded?

No. They’re two different obligations: transparency about the automated system comes from the AI Regulation, and information about recording comes from GDPR. If the agent records or transcribes, both are needed.

If the agent doesn’t disclose, is the responsibility mine or the provider’s?

In front of the customer and in front of the authority, the company providing the service is responsible. That’s why it’s best for the notice to come configured by default in the service you sign up for, and to be reviewed every time the agent’s script is changed.

Can I use a human-sounding voice if I disclose?

Yes. The rule doesn’t forbid the voice from being natural: it requires the person to know that whoever is answering is an automated system. Once that’s disclosed, voice quality is a customer experience decision.