Why chatbots make things up
A language model predicts likely words. Without the right facts in front of it, it can produce a confident answer that is wrong. This is the main risk of putting a general-purpose model on a business website.
Answers come from your information
Chatdek builds a knowledge base from what you write and from pages read on your own website. The assistant answers from that material. If the knowledge does not cover a question, it says so instead of guessing.
Prices come from rows, not memory
For prices, fees and other figures, the assistant looks the matching row up in a price list you maintain and states exactly what it says, with the date it is true from. A figure that is not in the rows or the knowledge is removed from the reply.
It asks you when it does not know
On a front-desk setup the assistant can send you a short request when a visitor asks something your information does not cover, bring your answer back to the visitor, and offer to remember it.
The code decides what happens
Which tool runs (a booking card, a contact form, a hand-over) is decided by rules on the server, not by the model. The model only writes the words, and it is told exactly what the tool did. A claim such as "I have booked it" is removed unless the booking really happened.
Every reply is checked
After the reply is written it is checked against what really happened. Claims of actions not taken, invented figures and phrases you banned are removed before the visitor sees them.
You can see and teach
The coaching view shows what each agent could not answer, what visitors marked unhelpful and where a reply had to be corrected. You teach the answer once and it applies from then on.