Why Enéh says “I don’t know” — and why that’s its most important feature

Every business owner who has tried an AI chatbot has the same fear, and it is a reasonable one: what if it tells a customer something that isn’t true? A price you don’t charge. Opening hours you don’t keep. A discount that never existed. The customer believes it, turns up, and the argument happens with you.

This fear is the right place to start, because it points at the single most important question to ask about any AI assistant: not how clever it sounds, but what it does when it doesn’t know.

Why language models make things up

A large language model is built to continue text plausibly. Ask it “how much is the premium package?” and the most plausible continuation is a number — whether or not it has one. That is not a bug in one product; it is how the technology works. Telling the model “please don’t make things up” helps, but it is a request, not a guarantee.

So the answer can’t only be a better prompt. It has to be structure: where the facts come from, what is checked before an answer leaves, and what happens when a fact is missing.

Three layers between the question and the answer

1. Approved knowledge, looked up — not remembered

Everything the assistant knows about your business comes from knowledge you approved: your website, your price list, your menu or catalogue, notes you added. A person reviews it with you before it goes live. Prices, opening hours, products and stock are stored as structured data and looked up for each question, rather than left to the model’s memory of a page it once read.

2. A check before anything is sent

Before an answer reaches the visitor, it is checked against your data. If it contains a price, a phone number or — for shops — a stock claim that doesn’t appear in your approved knowledge, that part is removed. The customer never sees a number you didn’t publish.

3. An honest “I don’t know”, and a person

When the answer isn’t in your knowledge, the assistant says so plainly and offers the next best thing: it takes the visitor’s contact details and passes the question to your team, who get notified straight away. The visitor leaves with a promise of a real answer, not a made-up one.

You can step in at any time

Every conversation is visible in your dashboard as it happens. If you see one that needs a human touch — a frustrated customer, a big order, a delicate question — you can take it over live and reply yourself. The visitor sees a colleague join the same chat, with nothing lost from before.

“I don’t know” is also data

Every question the assistant couldn’t answer is listed for you. That list is surprisingly valuable: it shows exactly what your customers want to know that your website doesn’t say. Add the answer once — or tell us and we’ll add it — and from then on the assistant knows it.

Test any chatbot in two minutes

Before you choose any AI assistant, ours included, ask it these three things:

1. The price of something you don’t sell.

2. Whether you are open on a day you are definitely closed.

3. A detail that is nowhere on your website — say, whether you offer a loyalty discount.

A trustworthy assistant answers the second correctly and says “I don’t have that information” to the first and third — then offers to connect you with a person. If it confidently invents an answer to any of them, it will one day do the same to a customer.

Try to catch it out

Every one of our industry demos runs with these checks. Pick one and ask it something it can’t know. Questions about how it works? The FAQ covers data, privacy and setup.

Magyarul olvasnád? Miért mondja az Enéh, hogy „nem tudom” — és miért ez a legfontosabb funkciója.