Skip to content

A website chatbot is only as good as what it actually knows

2 min readUpdated 17 September 2026

The short version

When a website assistant performs badly, the instinct is to blame the AI. Far more often it was deployed with nothing but the website's marketing copy to work from, so it cannot answer the questions customers actually ask — pricing ranges, service areas, lead times, what happens next. Fixing that is a business-knowledge problem, not a model problem.

Plenty of businesses have now tried a website chatbot and come away unimpressed. The usual verdict is that the AI was not good enough.

Occasionally that is true. Far more often, the assistant was deployed with nothing useful to work from.

Marketing copy does not answer buying questions

A typical chatbot is pointed at the website and left to it. But a website is written to persuade, not to inform in detail. It says the work is high quality and the team is experienced. It rarely says what a bathroom renovation generally costs, how far out you are booked, or whether you cover the Sunshine Coast.

Those are the questions people actually ask before enquiring. If the assistant does not know the answers, it does one of two things, and both are bad: it apologises repeatedly, or it invents something plausible.

The information exists — it is just not written down

Ask a business owner what customers ask most and you will get a fluent answer in under a minute. Rough price ranges. Where they will and will not travel. What happens after someone gets in touch. Which jobs they turn down.

Almost none of it is written anywhere. It lives in the owner's head and in years of sent email.

Getting that out and into a form an assistant can use is the actual work of building a good one. It is unglamorous, it is mostly conversation, and it matters far more than which model sits underneath.

Knowing when to stop is a feature

A confident wrong answer costs more than no answer. If a customer is told a job will take two weeks and it takes six, the assistant has created a problem a person now has to fix.

A well-built assistant is explicitly told what it must not commit to, and hands over instead. "I don't want to guess at that — let me get someone to confirm" protects the relationship and still captures the lead.

It needs an owner

Prices change. Services get added. You stop covering an area, or start.

An assistant given a snapshot of the business and then left alone drifts out of date exactly the way a website does, except it is speaking to customers directly and with apparent authority. It needs someone reviewing what it is saying, on a schedule.

That is the part most people skip, and it is why the second year of a chatbot is usually worse than the first. It is also the half of the monthly fee most easily left out of a quote: keeping an assistant switched on costs a few hundred dollars a month, and keeping it accurate contains somebody's time.

Key points

  • Most website chatbots are given only the public marketing copy, which is exactly the information that does not answer a buyer's real questions.
  • The questions that decide a sale — cost, timeline, service area, process — usually live in someone's head or an inbox, not on the website.
  • An assistant that says "I'm not sure, let me put you through to someone" is more valuable than one that guesses confidently and gets it wrong.
  • Knowledge goes stale, so an assistant needs an owner and a review schedule the same way a price list does.

Common questions

Will an AI assistant replace answering enquiries myself?

Not entirely, and it should not try. The useful split is that the assistant handles the repetitive qualifying questions immediately, at any hour, and hands over a warmer and better-informed enquiry to a person.

What if it gives a customer wrong information?

That is the real risk, and it is managed by controlling what the assistant knows and instructing it to hand over rather than speculate. Anything commercially binding — a firm quote, a delivery promise — should require a human.

How much of my business does it need to know?

Less than people expect. Services, rough pricing logic, service area, lead times, what happens after an enquiry, and the twenty questions you answer most often covers the majority of conversations.

Want a look at how your business is actually doing on this?

We will tell you what we find, whether or not it turns into work.

Not sure where you stand? Take the 2-minute website check