AI chatbots for small business: what they do well, what they get wrong

What an AI chatbot for small business can reliably do, where it makes things up, and how to set one up so customers get help instead of nonsense.

Written by Kevin MatthewsHead of Development

Explainer5 min read

Start reading
AI chatbot for small business: an electrician's website assistant answering from its own pages and passing price questions to a person
A good chatbot answers only from your own pages and hands prices and complaints to a person.

An AI chatbot for small business can answer the same ten questions your team gets every day, at 2am, without getting tired. It can also state something false in a friendly tone, which is the part nobody puts in the sales deck. This read explains what these chatbots actually are, where they shine, and where they let you down.

What is an AI chatbot, and how is it different from the old kind?

An AI chatbot is a chat window on your website or app that understands ordinary sentences and replies in ordinary sentences. The old kind of chatbot followed a fixed menu: press 1 for opening hours, press 2 for pricing. If you asked something it had not been told about, it gave up.

An AI chatbot is built on a large language model, the same kind of technology behind ChatGPT. It does not look up a script. It writes a reply each time, based on what the person asked and on the material you gave it. That makes it flexible, and it is also why it can go wrong in ways a menu never could.

What do AI chatbots do well?

AI chatbots do well at repeated questions with clear answers that already exist in your content. Think opening hours, delivery times, how to reschedule, what is included in a service, which product suits which need. They are fast, patient, and available outside business hours.

They also help in quieter ways. A chatbot can collect the details of an enquiry before a person steps in, so your team starts with the name, the problem and the contact details already in hand. It can point people to the right page instead of making them hunt. And it shows you, in plain text, what customers actually ask, which is useful for fixing your website.

Where do AI chatbots get things wrong?

AI chatbots get things wrong when the answer is not in the material they were given, or when the question needs judgement. The technical name for confidently inventing an answer is a hallucination. The chatbot is not lying on purpose. It is predicting plausible words, and a plausible answer is not always a true one.

Situation What can go wrong
A question your content does not cover It guesses and sounds sure about it
Prices, stock or availability It quotes old or wrong figures
Complaints and upset customers It replies in a tone that makes things worse
Anything medical, legal or financial It gives advice it is not qualified to give
Unusual or tricky requests It agrees to things your business never offered
Table 1Common ways an AI chatbot goes wrong, and the situation that triggers each.

Should your business use an AI chatbot?

Your business should use an AI chatbot if you answer the same questions repeatedly and the answers are written down somewhere. It is a poor fit if most enquiries are one-off, emotional or complicated, or if your information changes by the hour and nobody keeps it up to date.

Here is a quick test. Look at the last 50 messages your team received. If more than half were a version of questions you could answer from your website, a chatbot will probably save you time. If most needed a human to think, a better contact form and a faster reply from your team will serve you better, and cost less.

How do you set up a chatbot that does not embarrass you?

Set it up with a short, approved set of content, firm limits on what it may say, and an easy route to a person. The details matter more than the brand of AI underneath.

The setup that works

  • Approved content only. The chatbot answers from pages and documents you have reviewed. If the answer is not there, it says so.
  • A clear "I don't know". It should say it cannot help with that and offer to pass the question on, instead of guessing.
  • A visible hand-off. Every conversation has a way to reach a person: a button, a phone number or a form that goes straight to your inbox or your Slack channel.
  • Boundaries written in. It never quotes custom prices, promises outcomes, or gives medical, legal or financial advice.
  • A log you actually read. Review conversations weekly at first. You will spot gaps in your content within days.
  • A privacy check. Know where conversations are stored and who can see them before you switch it on.

A chatbot also works best when it connects to the rest of your business. Sending qualified enquiries into HubSpot, alerting your team in Slack or checking an order status in your store all turn a chat window into part of your workflow. If you would like a lead engineer to look at what would suit your setup, that is exactly the kind of work we do, and the plan is agreed in writing before anything is built.

Finished this read?

  1. OpenedToday
  2. FinishedNot yet

Written by

Kevin MatthewsHead of Development

Kevin leads engineering at SingletonTec and has spent 10+ years building production-grade websites, apps and systems for businesses.

Ask Kevin a questionMore from the lead engineer

// questions about this read

Still wondering?Start with these.

Short answers from Kevin and the team. Not here? Ask Kevin directly.

  1. They can be, if you answer the same questions again and again and those answers are already written down. They are less useful when your enquiries are mostly unique or sensitive. Check your last 50 customer messages: if over half are repeat questions, a chatbot is a reasonable bet.

    Talk to an engineer about thisHow much does AI automation cost in Australia?

  2. Technically yes, but pasting a general ChatGPT window onto your site is a poor idea. It will not know your prices, policies or limits, and it can invent answers about your business. A proper setup limits the chatbot to your approved content, adds a hand-off to a person, and checks how conversation data is handled.

    See custom software and automationWhat not to paste into ChatGPT: AI privacy for Australian businesses

  3. A hallucination is when an AI states something false with total confidence. You cannot remove the risk entirely, but you can shrink it a lot: restrict the chatbot to content you have approved, tell it to say "I don't know" when the answer is missing, and review real conversations regularly so you can fix gaps early.

    Talk to an engineer about this

  4. For a small business it is better to think of a chatbot as a first responder, not a replacement. It handles the simple repeat questions so your people have time for the conversations that need care and judgement. Keeping a real person easy to reach also protects your reputation when something goes wrong.

    AI for small business: where to start (and what to ignore)

  5. It can be, with care. Tell people what you are collecting and why, collect only what you need, and understand where the conversation is stored and who can see it. Avoid collecting sensitive details such as health or financial information through a general chatbot unless it has been built for that purpose.

    What not to paste into ChatGPT: AI privacy for Australian businessesBook a 10-min call