An AI chatbot on your website: when it pays off and when it doesn't
How a modern chatbot answers from your content and so can't make up a price, where it actually helps, when it's money thrown away and what to measure.
In short
- It answers from your own content, so it can't make up a price or a deadline. If it doesn't have the answer, it says so.
- Most of the setup work is preparing content, not programming.
- With three visitors a day it won't help; there the problem is traffic, not answering.
Website chatbots have a bad reputation, and rightly so. The first generation was a tree of pre-written answers: if your question wasn't one of the expected ones, you got "Sorry, I don't understand" and a button to a form. That helped no one.
Today's chatbots are different. They don't guess from general knowledge; they answer from your content: service pages, FAQs, instructions, email archives. Below is how it works, when it pays off and when it's money thrown away.
Where it gets its answers
This is the only technical part worth understanding, because everything else follows from it.
Your content is split into short paragraphs and stored in a database. When a visitor asks something, the system first finds the paragraphs closest to the question and only then builds an answer from them. The language model doesn't answer from memory; it answers from what it has just found.
The result: the chatbot can't make up a price or a deadline that isn't in your content. If it doesn't have the answer, it says so. That is exactly what separates a useful chatbot from a dangerous one.
The second result: quality depends on your content, not on the model. If your website doesn't say what the price includes, the chatbot won't be able to tell anyone either. That's why most of the setup work is writing and editing content, not programming.
What it actually does
It answers when you're not around. For a service business, a large share of website enquiries come in the evening and at weekends, when a form means waiting until Monday. A chatbot removes that wait.
It answers the same question for the fiftieth time. Opening hours, service area, what you need from the customer, how long it takes, whether you come out for a site visit. These are questions someone in the business is copying from email to email today.
It pre-qualifies. Before it hands the conversation to a person, it can ask what you need to know anyway: what kind of property, when, where. Your team gets an enquiry with context instead of a bare "Hello, I'd like a quote".
When it pays off
When you get a lot of repeat questions before a purchase. Services where customers decide based on terms, deadlines and scope: construction, transport, healthcare, education, equipment hire.
When your website gets decent traffic but converts poorly. If enough people visit and too few get in touch, the cause is often an unanswered question, not too little traffic. A chatbot targets exactly that gap.
When you have the content. A price list, common answers, service descriptions. If these exist, setup is quick.
When it doesn't pay off
When you get three visitors a day. Then the problem is traffic, and the money belongs in SEO or a website, not a chatbot.
When every customer is a separate case and there are no repeat questions. Custom project work can't be described in advance; there a chatbot is just a more expensive contact form.
When you have no one to take over the conversation when the chatbot hands it on. The handover has to be fast, or the experience is worse than having no chatbot at all.
What you should never leave to a chatbot
Binding quotes and prices for a project no one has seen. A chatbot can give a range and say what it depends on; a person makes the final quote.
Medical, legal and financial advice. Here the answer should point to a person, not the model.
Personal data without consent. If it collects contact details, it has to say who will process the data and why, just like a form. That isn't a detail; it's a GDPR requirement.
What to measure to know whether it works
Not the number of conversations. That number grows by itself and means nothing.
Measure three things: how many conversations end with contact details or a booking, how many end with "I can't answer that" (that's a list of content missing from your website), and how many come in outside working hours. The third number is the one that shows whether you really got something you didn't have before.
What it costs
Setup depends on how much content needs preparing and what the chatbot has to connect to: just the website, or also your calendar, CRM and stock. The running cost depends on the number of conversations and the chosen model, and for a small business it is usually small compared to the hours of work it saves.
Because both differ from project to project, we only give figures once we know what needs doing. Tell us what you're interested in and you'll get a detailed quote.
More about what we do in this area is on the AI solutions and automation page.
Sources
Found this useful? Add GudWeb as a preferred source on Google