If you run a small business with a website, you have almost certainly been offered an AI chatbot in the past year. The pitch is always the same. It answers questions around the clock, captures leads while you sleep, and costs less than hiring anyone. Some of that is true. Quite a lot of it depends on details nobody mentions in the sales email.

This is an honest look at when a chat widget is a sensible addition to a small business site, when it quietly makes things worse, and what it really costs once you include the work of keeping it accurate.

What has actually changed

Chat widgets are not new. Businesses have been bolting them onto websites for well over a decade. The older generation followed rigid decision trees: you picked from a short menu, the bot produced a canned reply, and if your question did not fit one of the predefined branches you were stuck. They were obviously robotic, and most visitors treated them accordingly.

The current generation is built on language models. It reads a question phrased however the visitor happens to phrase it, replies in natural language, and, when configured properly, draws its answers from your own content rather than a generic script. That is a genuine step change in capability. Visitors can ask something awkward and still get a sensible reply.

The capability being better is not the same as it being right for your business. Those are two separate questions, and the second one is the one worth spending time on.

Where a chatbot genuinely earns its place

There are clear situations where a well-configured chatbot pays for itself. The common thread is repetition and volume.

  • You answer the same handful of questions constantly. Opening hours, which areas you cover, roughly what things cost, how the booking process works. If your inbox is full of the same five questions, a bot handles them without anyone being interrupted.
  • Enquiries arrive outside working hours. Someone browsing at nine in the evening either gets an answer or waits until morning. A fair number will not wait, and a straightforward answer at the right moment can be the difference between an enquiry and a lost one.
  • You have a large catalogue or a wide service range. When a site has dozens of products or services, navigation alone often fails people. Describing what they need in their own words and being pointed to the right page is genuinely faster than hunting through a menu.
  • Routine questions eat meaningful staff time. This is the one that decides it. If handling basic queries takes hours out of someone's week, automating part of that has a measurable value. If it takes twenty minutes, it does not.

Where a chatbot actively hurts

This side deserves equal weight, because it is the half that rarely appears in the sales material. A chatbot is not a neutral addition. If it is the wrong fit, it costs you more than the subscription.

  • Low-traffic sites. If a few dozen people visit each week, a chat widget adds monthly cost and visual clutter for an audience that would have been perfectly happy with a contact form. There is no volume for it to save you from.
  • Businesses built on personal expertise. If people choose you because of judgement, experience and relationships, an obvious bot sitting in the corner of the page works against the impression you have spent years building. The first interaction sets expectations, and an automated one sets the wrong ones.
  • Bots that intercept people already on their way to contact you. A widget that pops up over someone heading for the contact page has added a step rather than removed one. They had already decided to get in touch, and you put a conversation in front of them instead.
  • Any setup where the bot cannot answer and simply loops. Being asked to rephrase a question three times, then offered the same unhelpful suggestion again, is more frustrating than having no bot at all. People leave with a worse impression than if the page had stayed quiet.

Warning

A looping bot is not a neutral failure. A visitor who cannot get an answer and cannot reach a person usually leaves the site entirely, and they rarely come back to try the contact form instead.

The honest cost picture

The headline price is usually a monthly subscription, and most providers scale that price with conversation volume. The quoted entry tier tends to assume modest usage, so a busy month can cost noticeably more than the figure you budgeted for. Worth checking where the tiers sit before you commit.

The larger cost is setup, and it applies even when the software itself is free. A chatbot is only as good as the information you give it. Somebody has to assemble accurate answers about your services, pricing, process and policies, then shape them into something the bot can use reliably. That is real work, and it is usually work only you or your team can do.

Then it has to be maintained. Prices change, services get retired, opening hours shift, a new location opens. A neglected chatbot confidently quoting last year's prices is worse than no chatbot, because now you are correcting an expectation you created yourself.

Tip

Before you sign up, decide who owns the chatbot's content and put a recurring reminder in the diary to review it. If nobody is clearly responsible for keeping it accurate, it will drift out of date within months.

The risk that matters most

Stated plainly, a language model can say something completely incorrect with complete confidence. It does not hesitate, it does not flag uncertainty, and it reads exactly like a correct answer. This is a known characteristic of the technology rather than a rare glitch.

For a small business that matters commercially. If your bot promises a service you do not offer, quotes a price you will not honour, or invents a delivery timescale, you own that conversation. The customer was told it by something on your website, and arguing the point afterwards is a poor start to a relationship.

The mitigations are straightforward but not optional. Constrain the bot to your actual published content rather than letting it answer from general knowledge. Be explicit about the subjects it should not attempt: quotes, legal or medical questions, anything involving a commitment. And give it an easy, immediate handover to a human when it reaches the edge of what it knows.

If you decide to go ahead

Assuming the fit looks right, a handful of decisions separate a chatbot that helps from one that quietly annoys people.

  1. Feed it your real content. Your FAQ and service pages are the most valuable source material you have. If those pages are thin or out of date, fix them first, because the bot will inherit every gap in them.
  2. Make the escalation path obvious. A visible route to a person or a contact form should be available at every point in the conversation, not buried behind two more attempts at an answer.
  3. Be transparent that it is automated. People work it out within a message or two, and they resent the pretence. Saying so up front costs nothing and sets fair expectations.
  4. Read the transcripts in the first few weeks. This is the single most informative thing you can do. You will find out what people actually ask, where the bot gets it wrong, and which pages on your site are failing to explain something. That insight is often worth more than the bot itself.
  5. Check the effect on page speed. Some widgets are heavy. Measure your pages before and after installing one, particularly on mobile, and make sure you have not traded a faster site for a feature most visitors never open. I test this as part of my site speed and web development work, because a widget that slows every page rarely pays for itself.

The alternative worth considering first

A genuinely good FAQ page solves much of the same problem, and it is consistently underrated. You pay no subscription, any number of visitors can read it at once, the answers are the ones you wrote rather than ones a model generated, and your pages stay exactly as light as they were.

It also does something a chatbot cannot. Written answers on a public page are indexed by search engines and readable by AI search systems, so they can bring people to your site rather than only serving those already on it. A bot answers one visitor at a time, privately, and leaves nothing behind.

If most of your enquiries are the same repeat questions, write the answers down publicly before you pay a bot to recite them. You may find the enquiry volume drops enough that the question answers itself.

Info

Writing a proper FAQ is also the groundwork for a chatbot. If you decide to add one later, that content becomes the source it draws on, so the effort is never wasted.

A simple test before you commit

Look back at your last fifty enquiries and sort them into two piles.

If most of them are routine questions that a clearly written page could have answered, your problem is the website, not the absence of a chatbot. Fix the pages first and the enquiries improve on their own. If most of them needed a real conversation, judgement or a tailored quote, a bot will not help much either, because those are exactly the exchanges it cannot handle.

A chatbot earns its place in the middle ground: enough volume of genuinely repetitive questions that automating them saves real time, and enough accurate content to answer them safely. That describes some small businesses. It does not describe most of them, whatever the sales email says.

I would rather fix the pages that answer your common questions before adding a widget that repeats them. If you are weighing one up, send me the project brief with your current site and the questions you field most often.