What your digital twin does when it doesn't know something

Because that is the moment that decides whether you can trust it. A digital twin works with what is in its brain: your materials, your rules and your way of explaining things. No brain holds everything, and questions from real clients always end up going beyond what was planned.

A language model, the technology underneath any twin, has a well-known tendency: when information is missing, it can produce an answer that sounds right but rests on nothing. In a casual conversation that is a mistake. In a conversation between a professional and a prospective client it is something else: a false statement about your price, your method or your timescales, made with your face and your voice.

A risk to trust and a legal risk. The trust risk is obvious: one invented answer is enough for a client to doubt all the others.

The legal risk is written down. In Spain, the Unfair Competition Act treats as misleading any false information, or true information presented in a way that misleads, about the nature of a service, its main features, its availability, its benefits or its price. Other European countries have equivalent rules. Whatever your twin tells a prospective client is your business speaking, under the same rules as your website or your brochure.

And if the answer is about a person, the GDPR comes in: Article 5 requires personal data to be accurate and inaccurate data to be corrected without delay. A twin that improvises facts about someone does not meet that principle.

The same as you would. When you do not know something in front of a client, you do not make it up: you admit the limit, offer the next step and make a note of it. The table compares that behaviour with what does not work.

The right-hand column has three parts that should always be there: admitting the limit clearly, offering a way forward and not leaving the client without an answer.

Separating the two is exactly what this topic calls for. The table sums it up.

What comes next closes the loop: every gap is noted and reaches you, so you can decide whether to cover it with new material or with a rule.

With three things already in your hands. First, complete materials on what you are asked most. Gaps tend to sit where you have never written anything down: prices for particular cases, timescales, what you do not do. Writing it down closes most of them.

Second, boundary rules in the Book of Criteria, each with its condition, its action, its reason and its source. For example: if a question cannot be answered from the materials, the twin does not guess and suggests a conversation with you. Or: if someone asks for a price that is not listed, the twin explains how it is worked out and leaves the figure to you. The Book works in versions: each change is activated with a note explaining what changes.

Third, a list of difficult questions with written answers: the ones that make you uncomfortable, the ones that lead to misunderstandings, the ones asked by clients who are not a good fit. With those answers written down, the brain has fewer gaps.

Treat them as information. Every question the twin cannot answer tells you something about your clients: what worries them, what they do not understand, what they expect from you. ISO 30401, on knowledge management systems, is a reminder that useful knowledge is maintained, reviewed and improved; a twin is no different.

A good rhythm is to review, from time to time, the questions that did not get a good answer and decide, one by one, whether they deserve new material, a new rule or neither, because they are not yours to answer.

That they are talking to an AI, and that the AI has limits. Article 50 of the EU AI Act requires them to be told clearly at the latest at the first interaction, and the twin says so in its first turn. Adding, on your website or in the appointment confirmation, that the twin will pass on to you whatever it cannot answer helps make sure nobody expects from it what it cannot give.

• Today: each twin has its own Master Brain with the materials you load, the Book of Criteria holds your boundary rules with their source, the AI notice is mandatory and revocation is immediate.
• In development: logging every unanswered question and notifying the professional, so no gap is lost.
• In development: the twin applying your rules in the conversation and citing the material or rule behind each answer.