Coweb·Hub
← All postsAI Development

What an AI chatbot on a website actually does

Coweb Hub · 17 November 2026 · 3 min read

Most articles about AI on websites are written by people selling AI on websites. Here is a plainer account, including where it does not help.

The actual job

The useful version is not a chatbot in the sense of a toy that answers trivia. It is the thing that stops a visitor leaving without an answer.

Somebody arrives at eleven at night with a specific question. Do you deliver to my area. Can you do this before Friday. What does it cost for a place my size. Nobody is awake. The site either answers or it does not.

That is the whole job. Answer the question, capture who asked it, and tell you in the morning.

What it handles well

Questions you already answer out loud every week. Scope, coverage, timelines, price ranges, what happens next, what you need from the customer before you can start.

Qualifying. A good assistant can work out whether somebody is a fit before it takes up your time, and say so politely when they are not.

Routing. Collecting the right details for the right person, so the message that reaches you is actionable rather than just a notification that someone exists.

Out of hours. This is where most of the value sits. It is not replacing your working day; it is covering the two thirds of the week you are not at your desk.

Where it fails

It fails when it is asked to know things nobody told it. An assistant is only as good as what it has been given about your business, and the failure mode is confident invention rather than silence.

This is the thing to get right. A well built assistant says it does not know, takes the question, and makes sure you see it. A badly built one makes something up, and you find out when a customer quotes it back to you.

It also fails at genuine negotiation, at anything emotionally delicate, and at judgment calls about an unusual job. Those should reach a person quickly, and the assistant should know it.

What it costs to run

Two numbers. The build, which varies enormously depending on whether it is a generic widget or something that actually knows your business and connects to your systems. And the running cost, which is usage based and for most small businesses is modest, typically single-digit to low double-digit dollars a month at small volumes.

Ask any supplier what the monthly running cost looks like at your expected volume. If they cannot answer, they have not thought about it, and you will find out from a bill.

When it is not worth it

If almost nobody visits your site, an assistant answers nobody. Fix the reason there is no traffic first.

If you can genuinely answer every enquiry within the hour yourself, the gain is smaller. Speed is most of the decision, and if you already have the speed, you already have the benefit.

If your work is so bespoke that every question needs judgment, an assistant that collects the right information and gets out of the way is more useful than one that attempts answers.

The question worth asking

The real question is not whether AI can answer your customers. It is whether anyone currently does, at the hour they ask.

If the honest answer is that enquiries sit until the morning, that is the gap. Whether you close it with software or a person is a separate decision, and both are legitimate.

Keep reading

1 comment

Comments from guests are checked before they appear. Logged-in comments post straight away.

One hub

8 services, one team

24-hour quote

On every inquiry

AI-native delivery

Built in, not bolted on

No lock-in

Project or retainer, your call