The assistant that says it is not a lawyer.

It is quarter past eleven at night. Someone who has just lost their mother opens a law firm’s website and types into the chat: “Should I accept the inheritance?” Tomorrow they have to talk to their sister, and they want to arrive with something clear.

There are two ways to fail that person. The first is not to answer until Monday. The second is to answer the question. An assistant that replies “it depends, but in general it is wise to accept with the benefit of inventory” is giving legal advice without knowing the case, without a conflict check and without a lawyer who answers for it.

The assistant at Orvalle, the fictional law firm I designed for this sector, does a third thing. It says it is an AI assistant and not a lawyer, explains that the question needs a person who knows the case and offers the first consultation at the right office. If the person wants to go on, it asks only for names to check for conflicts. All in the language they wrote in.

Saying it is a machine.

In the EU, Article 50 of the Artificial Intelligence Act has applied since 2 August 2026. An AI system that interacts directly with people must be designed so that they know they are dealing with a machine, unless that is obvious from the context. The Act also reaches those outside the EU when the system’s output is used in the Union, which is common for a Swiss firm with clients in Spain or Germany.

Switzerland does not yet have a specific AI law. But the Federal Data Protection and Information Commissioner considers that current data protection law applies directly to AI, and that people talking to such a system have the right to know whether they are dealing with a machine. By different routes, a firm in Lugano and one in Madrid arrive at the same first sentence.

On a law firm’s website, that sentence matters more than elsewhere. Clients may believe that what they type is already covered by professional secrecy, or that the answer binds the firm. Saying from the start who is speaking prevents both misunderstandings.

Secrecy, with a provider in between.

An AI assistant almost always runs on a model from an outside provider. That is nothing new for a law firm: email, cloud storage and practice management software also belong to third parties. What does not change is the duty of care. In Switzerland, Article 13 of the Lawyers Act requires lawyers to make their auxiliaries respect professional secrecy, and Article 321 of the Criminal Code also punishes auxiliaries who disclose it. In Spain, Article 22.4 of the General Statute requires lawyers to make collaborators, staff and anyone else who works with them respect it.

In practice, that means three things: knowing where the data is processed and whether it is stored, having a contract with the provider that says so in writing, and not putting into the assistant anything it does not need. The best confidential data for a chat is the data that never reaches it.

What the profession is already doing.

The legal profession is not late to AI. In January 2026, the General Council of the Spanish Legal Profession presented a White Paper on the use of AI in the profession, according to which 60% of practitioners already use applications with AI. The question is no longer whether a firm uses it, but where, within what limits and with what face towards the client.

How the Orvalle assistant behaves.

• It introduces itself. Its first sentence says it is the firm’s AI assistant and not a lawyer.
• It gives no legal advice. It explains how the firm works, what the first consultation costs and who handles each type of matter. Faced with a question about a case, it refers to a person.
• It does not ask for facts. If the client wants to go on, it asks for names for the conflict check, never for the story.
• It warns in time. If someone starts typing details of their case, it suggests keeping them for the consultation.
• It offers a person. There is always a visible way to the team, with each office’s hours and time zone.
• It speaks the client’s language. Italian, Spanish, German, French, English or Chinese, with the same rules in all of them.

Where the data goes.

An assistant like this does not need to keep conversations to work. It can delete the text when the session closes and keep only what the client chose to send: the names and the appointment. It can work without tracking cookies. And its knowledge base can be limited to what the firm has published, so that it does not invent services, prices or deadlines.

All of this is decided at the design stage, not afterwards. That is why, when I build an assistant for a business, I write down first what it must not do.

What the firm gains.

The assistant does not replace anyone at the firm. It does the work that is lost today: the late-night question nobody answers, the client abroad who does not dare to call, the five-minute call to explain what a first consultation costs. And it hands the first meeting names already checked and an appointment already paid for.

Write down the question you are asked most often in the first call, the one nobody should answer without knowing the case. Now imagine your website receives it at midnight, in another language. What would you rather happen: that the website answers, that it stays silent, or that it can explain why it does not answer and what to do next?

For me, in a law firm, only the third is acceptable, and it is the hardest to build. You can see the Orvalle assistant do it in the concept, in six languages. If you want one for your firm, start with the Strategic Diagnostic or, if your website is already well resolved, with the Automation Assessment.