Chatbot

Chatbots have been around for decades. The older kind follows a script: buttons or keywords lead down a decision tree to a canned reply. The newer kind runs on a large language model and can understand free-form questions and write its own replies. Both are called chatbots, and many real products mix the two.

A chatbot is narrower than an AI assistant. A chatbot on a dentist's website exists to answer that practice's questions and book appointments; a general assistant can help with almost any topic. The distinction matters because a narrow chatbot should refuse to wander off its job.

How it works

A scripted chatbot matches what the visitor clicks or types against a list of options and returns the matching reply. It cannot go wrong in surprising ways, but it cannot handle anything its designers did not foresee, and visitors quickly hit dead ends.

An AI chatbot sends the conversation to a language model together with hidden instructions, the system prompt, that set its role, tone and limits. To answer questions about a specific business, it usually looks up the relevant passages from that business's own documents before replying, so the answer is based on real policies rather than the model's general knowledge. This is called grounding.

Good chatbots also know when to stop. They hand over to a person when a question is outside their scope, when the customer is upset, or when an action such as a refund needs approval, and they pass the conversation history along so the customer does not repeat themselves. The quality of handover is often what customers remember, more than the answers themselves.

Example

A cleaning company adds a chatbot to its booking page. It is given the company's price sheet, service areas, cancellation policy and opening hours. A visitor asks whether the company cleans offices on Saturdays in a nearby town. The bot checks the service area and the hours, answers yes, and offers the booking form. When another visitor asks for a discount on a large contract, the bot collects their details and passes them to the owner.

Before launch, the owner writes thirty real questions from past emails, including awkward ones, and checks every answer. Two replies quoted a price that had changed, which revealed that the price sheet given to the bot was out of date. After a month, the owner reads the conversation logs every Friday. Questions the bot could not answer, such as whether the team brings its own products, become new lines in the source documents and new entries on the website FAQ, so the bot and the site improve together.

Why it matters for your business

A well-built chatbot answers routine questions at any hour and can capture leads that would otherwise leave the site. A badly built one invents policies, frustrates visitors and creates promises you may be held to. If you get only a handful of enquiries a week, a clear FAQ page and a fast contact form may serve you better than any bot.

Common mistakes

• Launching an AI chatbot without testing it on real customer questions, including difficult ones.
• Letting the bot answer from general knowledge instead of your own current documents.
• Hiding the way to reach a human, so frustrated visitors leave instead of asking.
• Not reviewing conversation logs, which are the best source of missing answers and new FAQ content.

Questions people ask

What is the difference between a chatbot and an AI assistant?

A chatbot is usually built for one job, such as answering a company's support questions, while an AI assistant is a general helper that can write, summarise, research and more. Many website chatbots today are powered by the same language models as assistants, but restricted to a narrow role.

Can a chatbot give wrong answers?

Yes. Scripted bots give wrong answers when their script is outdated; AI bots can also invent details that sound right. Limiting the bot to your own documents, testing it before launch and reviewing logs regularly reduces errors, but someone should still be responsible for what it tells customers.

Do I need a chatbot on my website?

Only if visitors ask the same questions often and you cannot answer them quickly yourself. For many small businesses, a clear FAQ, visible contact details and a short form do the job with less risk. A chatbot earns its place when enquiry volume is high or many questions arrive outside working hours.