AI chatbot
What is an AI chatbot?
An AI chatbot is a program on a website that answers visitors’ questions in ordinary language instead of sending them to a form or a phone number. A language model writes the answers, working from the company’s own material: product descriptions, delivery terms, opening hours. It takes the repetitive questions and hands anything outside them to a person.
Example
A spare-parts retailer gets around 400 messages a month. Four questions cover most of them: is the part in stock, how long is delivery, does it fit a given model, how do I send it back. A chatbot wired to the catalogue answers the first three from live data and, for the fourth, opens the returns page and offers a person.
With simple arithmetic: if 70% of 400 messages are settled without a person, about 120 reach the desk. At four minutes each, that is close to 19 hours a month. The hours do not disappear; they move to the conversations where a purchase is decided.
Answers from your own documents
A chatbot that invents does more harm than good. So the model does not answer from memory: it searches a prepared set of your texts and stays with what it finds, an approach known as retrieval-augmented generation. When the material holds no answer, the bot says so and passes the conversation on. That behaviour is configured on purpose — left alone, a language model would rather answer than admit it does not know.
Why it matters for a business
A chatbot earns its place when it is connected to real data and given a boundary. Stock from the warehouse system, delivery windows from the shipping terms, an order status once the customer identifies themselves: those answers are worth having. Vague ones are not. Transcripts are personal data — they are kept for a set period and deleted, and if an outside provider handles them, that belongs in a contract.
Since 2 August 2026 the EU regulation on artificial intelligence has required disclosure: a visitor should know they are talking to a machine rather than work it out halfway through. Where the bot cannot help, it should hand over as cleanly as a form would.
What to ask
- Which documents and systems does it read, and what does it do when there is no answer?
- How does a conversation reach a person, and during which hours?
- Where do the messages go, which provider processes them, and how long are they kept?
- How is the benefit measured — conversations closed without a person, not messages counted?
- Who updates the source texts when the terms change?
Connecting a chatbot to a warehouse system or to orders is part of building web software.