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

More and more companies put a virtual assistant on the website or on WhatsApp to answer outside hours. It is a reasonable idea, but it raises a practical question: does the customer have to know they are not talking to a person? This article gives the essentials in plain language. It is not a substitute for legal advice, and the rules may have moved on since we wrote it.

What the rule says

The European regulation on artificial intelligence has an article (Article 50) on transparency. According to the European Commission, that article has applied since 2 August 2026. In short, when a person interacts directly with an AI system, they must be told so, unless it is obvious from the context, and that exception is read narrowly. The information must be clear, and given right at the first interaction.

What this means in practice

• Say it is an AI, right at the start, in the channel where the conversation takes place, and not only on a terms page.
• Do not rely on "it's obvious". A friendly name and a photograph of a person can lead someone to think they are talking to a human.
• Use plain language: "I am a virtual assistant for [company]."
• Offer a way out: an easy way to talk to a person, especially for complaints, health or delicate matters.

An example of an opening

"Hello! I am a virtual assistant for [company] and can help with opening hours, bookings and simple questions. If you would rather talk to a person, write 'person' and we will pass your request to the team."

What a notice does not solve

• It does not prove everything is in order. Article 50 is one part of the rules; there are others, such as data protection.
• It does not correct wrong answers. If the assistant makes up prices or promises what it cannot, the notice does not avoid the problem. Limit what it can answer.
• It does not clarify your role. Whoever uses a supplier's tool and puts it at customers' service may have obligations of their own, and it is worth understanding which with the supplier and with legal support.

What to do, in order

1. Decide whether you really need an assistant or whether an honest automatic reply, like the one in the article answering outside hours, is enough.
2. Limit what it answers to a set of topics, with answers you control.
3. Say so in the channel where it talks.
4. Give a way out to a person and honour it.
5. Look after the data: what the person writes may contain personal information. Read ChatGPT and customer data.
6. Test with difficult cases before putting it in front of customers.

In process automation, each automation is tied to a concrete task and tested before it is switched on, and the first question in customer service is always whether the person knows who they are talking to. To choose what to automate, see which tasks to automate first.

Go through the channel as a customer

Open the assistant (or the automatic reply) as if you were a new customer. See how long passes before you realise you are not being served by a person, and whether any sentence from the assistant suggests otherwise. Ask it something it should not know, to see whether it makes things up. Write "I want to talk to a person" and see what happens. Every point where the experience misled you, or left you stuck, is one to fix before putting it in front of customers.

In short

• As a rule, whoever talks to an AI system should be told, clearly and from the start.
• According to the European Commission, Article 50 has applied since 2 August 2026.
• A friendly name and a photograph do not replace the notice.
• The notice does not solve everything: what the assistant says and the data it receives also need protecting.

Quick questions

Is it enough to put the notice in the terms and conditions?

No. The information must be clear in the channel itself, right at the start of the conversation, and not hidden on a page nobody reads.

What if the assistant only answers simple questions?

The exception for what is obvious is read narrowly. The safest course is always to say so, rather than depend on the person realising alone.

Does this also apply to emails or texts written with AI?

This article covers talking directly to an AI system, as in a chat. For other uses, such as texts written with AI help, the rules are different and are worth checking case by case.