This is the right question to ask, and the honest answer is that any language model can be wrong. What reduces it to a manageable level is grounding: the assistant is restricted to answering from your approved knowledge base rather than from general training, and it is instructed to say it does not know and hand over rather than guess. I then spend a full stage of the project trying to make it fail before launch, and thirty days after launch reviewing real conversations. Anyone who tells you their chatbot cannot hallucinate at all is selling you something.

Your content, not the open internet. That typically means your website pages, price lists, policy documents, product catalogue, past support emails and the answers currently living in your head that we capture during discovery. You approve the knowledge base before launch and can edit it afterwards without a developer.

One week for a focused FAQ assistant, two weeks for a qualifying assistant with CRM and booking integration, three to four weeks for multi-channel builds with custom integrations. The timeline depends far more on how quickly your content and approvals come back than on the build itself.

Yes, and it should. A conversation that ends in a transcript is worth very little. The assistant can book directly into your calendar, create or update a CRM record with tags and the full conversation attached, raise a support ticket, or look up an order — whichever your process needs.

It says so plainly and hands over, with the conversation history passed to whoever picks it up so the customer never has to repeat themselves. You set the escalation rules: certain topics, certain keywords, a set number of failed attempts, or an explicit request for a human. Refusing to guess is a feature, not a limitation.

Yes. The assistant replies in whatever language the customer opens with. For the languages that matter most to your business, I set approved terminology so your product names, service names and brand voice stay correct rather than being machine-translated into something slightly off.

Yes, always. Every assistant identifies itself as AI in its first message. This is partly ethics, partly law in a growing number of regions, and partly that it simply performs better — customers who know they are talking to an assistant ask clearer questions and are far less annoyed when it hands them to a human.

Possibly not. If you get a handful of enquiries a week and answer them quickly yourself, an assistant will not change your revenue and I will tell you that on the call. It earns its cost when you are fielding the same questions repeatedly, losing out-of-hours enquiries, or spending real time on calls that were never going to convert. I have talked clients out of this build before.