A wrong answer is worse than none

A woman asks the chat on a bakery's website: “Do you make a gluten-free sourdough, and is it baked on the same surface as the rest?” The bot answers at once, polite and certain: “Yes, we offer a gluten-free sourdough, baked in a separate area.” She buys it for her daughter, who cannot have gluten. The bakery has never baked anything gluten-free.

It was untrue, but not a deliberate lie. The bot was simply built to answer, and when the answer did not exist it produced one that sounded right.

Why bots make things up

General-purpose chatbots are trained to produce text that sounds convincing, not text that is necessarily correct. When a question touches something they do not know, a price, opening hours, whether a particular product exists, they fill the gap with a guess that looks like a fact. The problem is not that they lie. It is that they cannot tell the difference between what they know and what they are guessing.

For a chatbot meant to speak on behalf of a business, that is the most dangerous trait of all. A wrong fact, delivered with confidence, is persuasive. The difference between such a chatbot and Anna we take up in a separate article.

A confident wrong answer costs more than silence

An unanswered message is a problem, but at least you know it is there. A wrong answer is worse, because it looks like a solution. The customer trusts it, acts on it, and discovers nothing until she is standing at the counter, in the wrong place, or with the wrong expectations. At that point it is no longer about a lost sale. It is about trust that is hard to win back.

And the bill lands on your staff. They have to correct it, apologize, and fix something they never knew had gone wrong.

The rule that changes everything: answer only from your own knowledge

There is another way. Instead of letting the chatbot answer everything from some general well of knowledge, you let it answer only from what the business actually has: the website, the documents, the frequently asked questions, the product list, the opening hours. What is written there, it answers. What is written nowhere, it does not guess at.

This sounds simple, but it inverts the priorities. The goal is not to always answer. It is to answer correctly, and to stay silent rather than invent.

“I don't know” is also a good answer

A good answer is not always an answer. Sometimes the best one is to admit you do not know and pass the matter to a person. The chatbot then gathers the right details, writes your team a clean summary in their language, and tells the customer that someone is taking over. Nobody is left in a vacuum.

And if someone on the team is available, they can step straight into the conversation, take over, answer directly, and hand the baton back when they are done. The customer sees that a person has taken over, under their own name, and the conversation carries on in the same place.

You decide what she knows

Answers from your knowledge are only as good as the knowledge itself, which is exactly why you hold the reins. You review what she learned from the website and approve it. You see an overview of which topics she is ready to answer and where there are gaps. And when an answer is not good enough, you correct it once, approve the correction, and it holds from then on.

And you don't only teach her with documents. In the dashboard you can talk to Anna directly and teach her in conversation. You tell her what is missing, just as you would guide a new employee. What you teach her there becomes part of her knowledge from then on.

That way certainty is not an accident. It is a decision.

Frequently asked questions

Can an AI answer without making things up? Yes, if it is set up to answer only from the business's knowledge and to admit when the answer is not there. The problem is not the technology itself, but whether it is allowed to guess.

What happens when it does not know the answer? The right response is a handoff: it gathers the details, sends the team a summary, and tells the customer a person is taking over. Silence or a guess are the wrong answers.

Who is responsible for what it says? The business is. That is why it matters that you review and approve what she learns, and can correct her when needed. We go into this in full in who is responsible for what the AI says.

This is why Anna answers the way she does

Anna answers only from your own knowledge. She draws her answers from your website, your documents, your FAQs, and your products, and when the answer is nowhere to be found she says she does not know and passes the matter on with a summary, rather than making something up. You review what she learns, correct her once when needed, and trust that the correction holds.

A fast answer is worth little if it is wrong. Anna answers right away, but she answers correctly, and she would rather stay quiet than guess.

See how Anna works or try the chat right now, no signup needed. And if you missed the first article: why night is high season for inquiries.

← Allar greinar