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

The short answer

There are two answers, and only one of them travels.

The legal answer depends on the country you operate in, and it is changing fast enough that anything written today needs checking against your own jurisdiction. In Argentina, where we looked at the rules closely, no rule currently in force requires you to disclose that an AI is answering in customer service.

The commercial answer does travel, and it is the stronger of the two: 87% of customers say being able to reach a human is essential when a company uses generative AI. Hiding the AI is what gets punished, not using it.

What the legal picture looks like in one country

Argentina is worth walking through because it shows the shape the question usually takes.

Piece	What it is	What it requires
Personal Data Protection Act 25.326 (2000)	Data protection law	How you handle customer data. Its article 20 limits automated decisions, but does not require declaring the AI
The data authority's responsible-AI guide	A recommendation, built on transparency and explainability	Nothing binding
Reform bills in congress	Proposals drafted from the authority's own outline	Nothing yet: they are proposals

The closest the law comes points somewhere else entirely. Article 20 limits decisions taken solely by automated processing of data, and requires that a person can intervene.

In other words, the law is not concerned with whether your customer knows an AI replied. It is concerned with a machine deciding something that affects them, on its own. If your AI answers enquiries and hands off, you are far from that scenario. If it also resolved a cancellation or a rejection by itself, that is worth a lawyer's time.

This is a worked example, not advice for your market. If you operate in the EU, the US or anywhere else, the analysis is different and this post does not cover it.

So why disclose anyway

Because your customer already suspects, and the AI is not what bothers them.

Gartner surveyed 3,566 B2B and B2C customers in February and March 2026. 87% say it is essential to be able to reach a human when a company uses generative AI in service. In the same survey, 50% say their interactions are easier when the company uses AI.

Read together, those two numbers say something precise: AI does not annoy people. AI with no way out does.

And when you do not disclose it, the customer spends their first three messages working out what they are talking to instead of telling you their problem. That conversation gets longer, not shorter.

How to disclose without sounding like a legal disclaimer

The form matters more than the fact. Three rules.

Up front, in one line. Not in a block of terms. "Hi, I'm the assistant for [your business]. If you need someone from the team, just ask and I'll pass you over."

With the exit in the same sentence. Disclosing without offering a way out is the worst combination: you confirm the suspicion and resolve nothing.

Without apologising. An AI that opens by apologising for existing teaches the customer they got the second-tier service.

What disclosure alone does not cover

Disclosure is the floor. What holds trust up is the customer being able to leave whenever they want.

At a building management company we serve, close to four in ten conversations end up with a person. That proportion is the exit being used. If your handoff rate is zero, you do not have a very good AI: you have a closed door.

On your team's side, it also pays for the record of what the AI did to stay in the chat. When a customer disputes something, the conversation has to be readable end to end.

What to write into your data policy

This part is real wherever you operate, because it is about data rather than disclosure.

• What customer data from the conversation gets stored.
• Who processes it, and whether a third party is involved.
• How long it is kept.
• How the customer requests access or deletion.

None of that depends on whether a human or an AI did the answering. But adding AI is usually the moment somebody reviews it for the first time.

Key takeaways

• Whether disclosure is required depends on your jurisdiction. In Argentina, no rule in force requires it today.
• The closest Argentine law comes is article 20, which limits fully automated decisions rather than requiring disclosure.
• 87% of customers say reaching a human is essential. That is the reason that holds everywhere.
• Disclose in one line, up front, and offer the exit in the same sentence.
• The data-policy obligations are real regardless, and they are about data, not disclosure.